modelId
stringlengths 5
139
| author
stringlengths 2
42
| last_modified
timestamp[us, tz=UTC]date 2020-02-15 11:33:14
2025-09-06 00:36:47
| downloads
int64 0
223M
| likes
int64 0
11.7k
| library_name
stringclasses 540
values | tags
listlengths 1
4.05k
| pipeline_tag
stringclasses 55
values | createdAt
timestamp[us, tz=UTC]date 2022-03-02 23:29:04
2025-09-06 00:36:27
| card
stringlengths 11
1.01M
|
|---|---|---|---|---|---|---|---|---|---|
Thireus/DeepSeek-V3.1-THIREUS-IQ4_KS-SPECIAL_SPLIT
|
Thireus
| 2025-09-05T22:11:27Z | 4 | 0 | null |
[
"gguf",
"arxiv:2505.23786",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-08-25T12:42:16Z |
---
license: mit
---
# DeepSeek-V3.1
## 🤔 What is this [HuggingFace repository](https://huggingface.co/Thireus/DeepSeek-V3.1-THIREUS-BF16-SPECIAL_SPLIT/) about?
This repository provides **GGUF-quantized tensors** for the DeepSeek-V3.1 model (official repo: https://huggingface.co/deepseek-ai/DeepSeek-V3.1). These GGUF shards are designed to be used with **Thireus’ GGUF Tool Suite** (https://gguf.thireus.com), a collection of tools that automatically finds the perplexity-optimal mix of quantizations for any given VRAM and RAM target. With the Tool Suite, you can generate and download custom quantization “recipes” effortlessly.
- 📖 Read more: https://github.com/Thireus/GGUF-Tool-Suite
- 🔍 Example quant mixes: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
- 🛠️ Create your own recipe: https://colab.research.google.com/github/Thireus/GGUF-Tool-Suite/blob/main/quant_recipe_pipeline.ipynb
- 📂 Browse available quant shards: https://huggingface.co/Thireus/collections
*tl;dr: Expand the details section below*
<details>
```
cd ~
# Make sure to install all ik_llama.cpp compilation dependencies...
apt install python3-dev python3-pip python3-venv python3-wheel python3-setuptools git acl netcat-openbsd cmake # pipx
# Obtain ik_llama's Thireus version - Windows builds available at https://github.com/Thireus/ik_llama.cpp/releases
git clone https://github.com/Thireus/ik_llama.cpp
cd ik_llama.cpp
git pull
# Build ik_llama.cpp
cmake -B build -DGGML_AVX=ON -DGGML_AVX2=ON -DLLAMA_CURL=OFF -DGGML_MAX_CONTEXTS=2048
cmake --build build --config Release -j16
cd ..
# Obtain Thireus' GGUF-Tool-Suite
git clone https://github.com/Thireus/GGUF-Tool-Suite
# Download model quant mix from recipe file:
cd GGUF-Tool-Suite
rm -f download.conf # Make sure to copy the relevant download.conf for the model before running quant_assign.py
cp -f models/DeepSeek-R1-0528/download.conf . # Use the download.conf of the chosen model
mkdir -p kitchen && cd kitchen
../quant_downloader.sh ../recipe_examples/ik_harmonized_recipes/DeepSeek-R1-0528.ROOT-2.7921bpw-3.4451ppl.218GB-GGUF_14GB-GPU_204GB-CPU.90e3c2f_6f5170d.recipe
# Other recipe examples can be found at https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
# Launch ik_llama's llama-cli:
ulimit -n 99999 # Lifts "too many open files" limitation on Linux
~/ik_llama.cpp/build/bin/llama-cli \
-m DeepSeek-R1-0528-THIREUS-BF16-SPECIAL_TENSOR-00001-of-01148.gguf \
-mla 3 -fa -amb 512 -fmoe -ctk f16 -c 4096 -ngl 99 \
-ot "blk\.(3|4|5|6)\.ffn_.*=CUDA0" \
-ot "blk\.(7|8|9|10)\.ffn_.*=CUDA1" \
-ot exps=CPU -b 2048 -ub 1024 --warmup-batch --no-mmap --threads 36 \
--main-gpu 0 \
-p '<｜begin▁of▁sentence｜><｜User｜>What is the solution of x+5=-2?<｜Assistant｜><think>\n'
```
</details>
---
## ❓ Why does this Tool Suite exist?
1. **Compatibility & Speed** – [unsloth](https://huggingface.co/unsloth)’s dynamic quants may not always work optimally with `ik_llama.cpp`.
2. **Custom Rig Fit** – No off-the-shelf GGUF model perfectly matched my VRAM/RAM setup, so I built a way to tailor models and leverage extra VRAM/RAM to reduce perplexity.
3. **Automated PPL-Optimal Quantization** – To my knowledge, there was no open source flexible, automated method to minimize perplexity for any bits-per-weight (bpw) target—so I created one with excellent results!
---
## 📊 How does it compare to other GGUFs?
Here’s how DeepSeek-R1-0528 quantized with **Thireus’ GGUF Tool Suite** stacks up against other quantizers (lower perplexity = better at equal or lower bpw):
> _Note: The `recipe_examples` files illustrate good recipes. The Tool Suite computes the optimal ppl/bpw curve for you — just specify your target RAM, VRAM, and quant types, and `quant_assign.py` finds the best mix._
More perplexity/bpw graphs for other supported models: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/ppl_graphs
*All PPL values are computed with the parameters `-ctk f16 -c 512 -b 4096 -ub 4096`. Changing any of these parameters will alter the PPL. In particular, reducing `-b 4096 -ub 4096` increases the PPL, while increasing them decreases the PPL.*
---
## 🚀 How do I get started?
Check out the [GGUF Tool Suite README](https://github.com/Thireus/GGUF-Tool-Suite) — focus on these sections:
1. ⚠️ **Requirements** – Which `ik_llama.cpp` (or `llama.cpp`) version to use and how to compile.
- Windows binaries (no patching needed) at: https://github.com/Thireus/ik_llama.cpp/releases
2. 📥 **Download Model Shards** – Use `quant_downloader.sh` to fetch GGUF shards from any recipe.
- Recipe examples: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
3. 🧠 **Run a Downloaded Model** – Sample usage with `llama-cli`.
4. 🛠️ **Generate a Custom Recipe** – Produce recipes tailored to your VRAM/RAM target usage for optimum perplexity.
---
## ✅ Supported Models
Supported models are listed under `models/` in the [Tool Suite Github repo](https://github.com/Thireus/GGUF-Tool-Suite/tree/main/models). Presence of `ppl_results.csv` indicates official support and compatibility with `quant_assign.py`.
---
## 🤷♂️ Will I release baked dynamic quant GGUFs?
No, because I believe in **tailored quantization** for each user’s hardware. If you prefer ready-made shards, you are welcome to merge them via `llama-gguf-split --merge`, or request someone to publish them, or rely on generic GGUF dynamic quants such as [unsloth](https://huggingface.co/unsloth)'s.
Instead, I prefer to share examples of recipes so users can see exactly how they were produced (command included inside these recipe files) and tweak them for their own rigs. The `quant_downloader.sh` script handles automatic fetching and verification of each shard. Note that recipes provided by [Ubergarm](https://huggingface.co/ubergarm) on his model cards are also compatible with `quant_downloader.sh`.
Users who don’t trust the GGUF shards on HuggingFace can also quantize their own by passing recipe lines to `llama-quantize --custom-q` ([see example](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/models/DeepSeek-R1-0528/DeepSeek-R1-0528-THIREUS-ANY-SPECIAL.sh#L482-L486)). Run `llama-quantize --help` to list compatible quants for `quant_assign.py`. This approach is especially useful if you prefer `llama.cpp` over `ik_llama.cpp`.
---
## 📦 What’s in this repository?
- **00001 GGUF header shard** – Contains metadata (tokens, chat template, tensor count, etc.). This metadata can be explored directly from the HuggingFace web interface after clicking on that shard.
- **Tensor shards** – Each shard holds one tensor; see `tensors.map` for names, quant types, sizes, SHA-256 hash, shard IDs, etc.
- **GPG-signed files** – `tensors.map` and header shard are signed with the key in [trusted-keys.asc](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/trusted-keys.asc) for tamper detection.
- **Security note** – Some papers about various ways to attack GGUFs and LLMs are available online, such as https://arxiv.org/abs/2505.23786, and there are also more classic security exploits like CVE-2024-23496 and CVE-2024-25664 through CVE-2024-25668. Only use GGUFs from reputable, trusted authors—or alternatively self-quantize—to avoid potential exploits.
---
## 💡 Pro Tips
You can easily download the BF16 model version to quantize your own shards:
```
mkdir kitchen
echo '.*=bf16' > kitchen/bf16.recipe
cd kitchen
../quant_downloader.sh bf16.recipe
```
Enjoy optimized quantization! 🎉
|
bah63843/blockassist-bc-plump_fast_antelope_1757110079
|
bah63843
| 2025-09-05T22:08:50Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"plump fast antelope",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T22:08:42Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- plump fast antelope
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
qgallouedec/Qwen3-8B-SFT-20250905191103
|
qgallouedec
| 2025-09-05T22:08:13Z | 0 | 0 |
transformers
|
[
"transformers",
"safetensors",
"generated_from_trainer",
"trl",
"sft",
"hf_jobs",
"dataset:trl-lib/Capybara",
"base_model:Qwen/Qwen3-8B",
"base_model:finetune:Qwen/Qwen3-8B",
"endpoints_compatible",
"region:us"
] | null | 2025-09-05T19:11:58Z |
---
base_model: Qwen/Qwen3-8B
datasets: trl-lib/Capybara
library_name: transformers
model_name: Qwen3-8B-SFT-20250905191103
tags:
- generated_from_trainer
- trl
- sft
- hf_jobs
licence: license
---
# Model Card for Qwen3-8B-SFT-20250905191103
This model is a fine-tuned version of [Qwen/Qwen3-8B](https://huggingface.co/Qwen/Qwen3-8B) on the [trl-lib/Capybara](https://huggingface.co/datasets/trl-lib/Capybara) dataset.
It has been trained using [TRL](https://github.com/huggingface/trl).
## Quick start
```python
from transformers import pipeline
question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
generator = pipeline("text-generation", model="qgallouedec/Qwen3-8B-SFT-20250905191103", device="cuda")
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
print(output["generated_text"])
```
## Training procedure
This model was trained with SFT.
### Framework versions
- TRL: 0.23.0.dev0
- Transformers: 4.56.1
- Pytorch: 2.8.0+cu128
- Datasets: 4.0.0
- Tokenizers: 0.22.0
## Citations
Cite TRL as:
```bibtex
@misc{vonwerra2022trl,
title = {{TRL: Transformer Reinforcement Learning}},
author = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallou{\'e}dec},
year = 2020,
journal = {GitHub repository},
publisher = {GitHub},
howpublished = {\url{https://github.com/huggingface/trl}}
}
```
|
exala/db_aca2_12.1.1
|
exala
| 2025-09-05T22:07:59Z | 0 | 0 |
transformers
|
[
"transformers",
"safetensors",
"distilbert",
"text-classification",
"arxiv:1910.09700",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] |
text-classification
| 2025-09-05T22:07:45Z |
---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed]
|
Kaori1707/gemma-3-12b-it-r16
|
Kaori1707
| 2025-09-05T22:07:48Z | 0 | 0 |
transformers
|
[
"transformers",
"safetensors",
"generated_from_trainer",
"sft",
"trl",
"base_model:google/gemma-3-12b-it",
"base_model:finetune:google/gemma-3-12b-it",
"endpoints_compatible",
"region:us"
] | null | 2025-09-05T15:43:55Z |
---
base_model: google/gemma-3-12b-it
library_name: transformers
model_name: gemma-3-12b-it-r16
tags:
- generated_from_trainer
- sft
- trl
licence: license
---
# Model Card for gemma-3-12b-it-r16
This model is a fine-tuned version of [google/gemma-3-12b-it](https://huggingface.co/google/gemma-3-12b-it).
It has been trained using [TRL](https://github.com/huggingface/trl).
## Quick start
```python
from transformers import pipeline
question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
generator = pipeline("text-generation", model="Kaori1707/gemma-3-12b-it-r16", device="cuda")
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
print(output["generated_text"])
```
## Training procedure
This model was trained with SFT.
### Framework versions
- TRL: 0.19.1
- Transformers: 4.52.4
- Pytorch: 2.6.0
- Datasets: 4.0.0
- Tokenizers: 0.21.4
## Citations
Cite TRL as:
```bibtex
@misc{vonwerra2022trl,
title = {{TRL: Transformer Reinforcement Learning}},
author = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallou{\'e}dec},
year = 2020,
journal = {GitHub repository},
publisher = {GitHub},
howpublished = {\url{https://github.com/huggingface/trl}}
}
```
|
somrima0907/codeT5_model
|
somrima0907
| 2025-09-05T22:07:30Z | 0 | 0 | null |
[
"safetensors",
"t5",
"license:apache-2.0",
"region:us"
] | null | 2025-09-05T21:56:34Z |
---
license: apache-2.0
---
|
WijewardhanaNT/xnli_en_1000_3
|
WijewardhanaNT
| 2025-09-05T22:06:00Z | 0 | 0 |
transformers
|
[
"transformers",
"safetensors",
"arxiv:1910.09700",
"endpoints_compatible",
"region:us"
] | null | 2025-09-05T02:27:01Z |
---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed]
|
proshantasaha/gemma-3-1b-medical-finetuned
|
proshantasaha
| 2025-09-05T22:05:49Z | 20 | 0 |
transformers
|
[
"transformers",
"safetensors",
"gemma3_text",
"text-generation",
"conversational",
"arxiv:1910.09700",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] |
text-generation
| 2025-09-02T20:49:02Z |
---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed]
|
inferencerlabs/Kimi-K2-Instruct-MLX-3.985bit
|
inferencerlabs
| 2025-09-05T22:05:38Z | 2,007 | 4 |
mlx
|
[
"mlx",
"safetensors",
"kimi_k2",
"text-generation",
"conversational",
"custom_code",
"base_model:moonshotai/Kimi-K2-Instruct",
"base_model:quantized:moonshotai/Kimi-K2-Instruct",
"license:other",
"4-bit",
"region:us"
] |
text-generation
| 2025-07-26T07:01:13Z |
---
license: other
license_name: modified-mit
library_name: mlx
base_model: moonshotai/Kimi-K2-Instruct
pipeline_tag: text-generation
tags:
- mlx
---
**See Kimi-K2 Dynamic MLX in action - [https://youtu.be/-zfUvA2CDqE](https://youtu.be/-zfUvA2CDqE)**
*q3.985bit dynamic quant typically achieves 1.243 perplexity in our testing, slotting closer to q4 perplexity (1.168) than q3 perplexity (1.900).*
| Quantization | Perplexity |
|:------------:|:----------:|
| **q2** | 41.293 |
| **q3** | 1.900 |
| **q3.985** | 1.243 |
| **q4** | 1.168 |
| **q6** | 1.128 |
| **q8** | 1.128 |
## Usage Notes
* Runs on a single M3 Ultra 512GB RAM using [Inferencer app](https://inferencer.com)
* Requires expanding VRAM limit to at least ~500000 MB
* For a larger context window, 507000 is used in VRAM limit command below.
* `sudo sysctl iogpu.wired_limit_mb=507000`
* Expect ~20 tokens/s
* Quantized with a modified version of [MLX](https://github.com/ml-explore/mlx) 0.26
* For more details see [demonstration video](https://youtu.be/-zfUvA2CDqE) or visit [Kimi K2](https://moonshotai.github.io/Kimi-K2/).
|
weecology/cropmodel-deadtrees
|
weecology
| 2025-09-05T22:04:49Z | 0 | 2 | null |
[
"safetensors",
"model_hub_mixin",
"pytorch_model_hub_mixin",
"region:us"
] | null | 2024-08-26T18:10:02Z |
---
tags:
- model_hub_mixin
- pytorch_model_hub_mixin
---
This model has been pushed to the Hub using the [PytorchModelHubMixin](https://huggingface.co/docs/huggingface_hub/package_reference/mixins#huggingface_hub.PyTorchModelHubMixin) integration:
- Code: [More Information Needed]
- Paper: [More Information Needed]
- Docs: [More Information Needed]
|
mowen222/task-13-Qwen-Qwen2.5-3B-Instruct
|
mowen222
| 2025-09-05T22:04:24Z | 115 | 0 |
peft
|
[
"peft",
"safetensors",
"arxiv:1910.09700",
"base_model:Qwen/Qwen2.5-3B-Instruct",
"base_model:adapter:Qwen/Qwen2.5-3B-Instruct",
"region:us"
] | null | 2025-08-10T01:12:35Z |
---
base_model: Qwen/Qwen2.5-3B-Instruct
library_name: peft
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed]
### Framework versions
- PEFT 0.15.2
|
cwayneconnor/blockassist-bc-mute_loud_lynx_1757109601
|
cwayneconnor
| 2025-09-05T22:03:45Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"mute loud lynx",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T22:01:13Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- mute loud lynx
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
helmutsukocok/blockassist-bc-loud_scavenging_kangaroo_1757108285
|
helmutsukocok
| 2025-09-05T22:03:18Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"loud scavenging kangaroo",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T22:03:14Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- loud scavenging kangaroo
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-IQ2_BN_R4-SPECIAL_SPLIT
|
Thireus
| 2025-09-05T22:03:01Z | 48 | 0 | null |
[
"gguf",
"arxiv:2505.23786",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-07-21T09:33:11Z |
---
license: mit
---
# DeepSeek-TNG-R1T2-Chimera
## 🤔 What is this [HuggingFace repository](https://huggingface.co/Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_SPLIT/) about?
This repository provides **GGUF-quantized tensors** for the DeepSeek-TNG-R1T2-Chimera model (official repo: https://huggingface.co/tngtech/DeepSeek-TNG-R1T2-Chimera). These GGUF shards are designed to be used with **Thireus’ GGUF Tool Suite** (https://gguf.thireus.com), a collection of tools that automatically finds the perplexity-optimal mix of quantizations for any given VRAM and RAM target. With the Tool Suite, you can generate and download custom quantization “recipes” effortlessly.
- 📖 Read more: https://github.com/Thireus/GGUF-Tool-Suite
- 🔍 Example quant mixes: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
- 🛠️ Create your own recipe: https://colab.research.google.com/github/Thireus/GGUF-Tool-Suite/blob/main/quant_recipe_pipeline.ipynb
- 📂 Browse available quant shards: https://huggingface.co/Thireus/collections
*tl;dr: Expand the details section below*
<details>
```
cd ~
# Make sure to install all ik_llama.cpp compilation dependencies...
apt install python3-dev python3-pip python3-venv python3-wheel python3-setuptools git acl netcat-openbsd cmake # pipx
# Obtain ik_llama's Thireus version - Windows builds available at https://github.com/Thireus/ik_llama.cpp/releases
git clone https://github.com/Thireus/ik_llama.cpp
cd ik_llama.cpp
git pull
# Build ik_llama.cpp
cmake -B build -DGGML_AVX=ON -DGGML_AVX2=ON -DLLAMA_CURL=OFF -DGGML_MAX_CONTEXTS=2048
cmake --build build --config Release -j16
cd ..
# Obtain Thireus' GGUF-Tool-Suite
git clone https://github.com/Thireus/GGUF-Tool-Suite
# Download model quant mix from recipe file:
cd GGUF-Tool-Suite
rm -f download.conf # Make sure to copy the relevant download.conf for the model before running quant_assign.py
cp -f models/DeepSeek-TNG-R1T2-Chimera/download.conf . # Use the download.conf of the chosen model
mkdir -p kitchen && cd kitchen
../quant_downloader.sh ../recipe_examples/ik_llama.cpp_recipes/DeepSeek-TNG-R1T2-Chimera.ROOT-3.0624bpw-3.3657ppl.238GB-GGUF_11GB-GPU_227GB-CPU.13549e6_1ac857a.recipe
# Launch ik_llama's llama-cli:
ulimit -n 9999 # Lifts "too many open files" limitation on Linux
~/ik_llama.cpp/build/bin/llama-cli \
-m DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_TENSOR-00001-of-01148.gguf \
-mla 3 -fa -amb 512 -fmoe -ctk f16 -c 4096 -ngl 99 \
-ot "blk\.(3|4|5|6)\.ffn_.*=CUDA0" \
-ot "blk\.(7|8|9|10)\.ffn_.*=CUDA1" \
-ot exps=CPU -b 2048 -ub 1024 --warmup-batch --no-mmap --threads 36 \
--main-gpu 0 \
-p '<｜begin▁of▁sentence｜><｜User｜>What is the solution of x+5=-2?<｜Assistant｜><think>\n'
```
</details>
---
## ❓ Why does this Tool Suite exist?
1. **Compatibility & Speed** – [unsloth](https://huggingface.co/unsloth)’s dynamic quants may not always work optimally with `ik_llama.cpp`.
2. **Custom Rig Fit** – No off-the-shelf GGUF model perfectly matched my VRAM/RAM setup, so I built a way to tailor models and leverage extra VRAM/RAM to reduce perplexity.
3. **Automated PPL-Optimal Quantization** – To my knowledge, there was no flexible, automated method to minimize perplexity for any bits-per-weight (bpw) target—so I created one with excellent results!
---
## 📊 How does it compare to other GGUFs?
Here’s how DeepSeek-R1-0528 quantized with **Thireus’ GGUF Tool Suite** stacks up against other quantizers (lower perplexity = better at equal or lower bpw):
> _Note: The `recipe_examples` files illustrate good recipes. The Tool Suite computes the optimal ppl/bpw curve for you — just specify your target RAM, VRAM, and quant types, and `quant_assign.py` finds the best mix._
More perplexity/bpw graphs for other supported models: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/ppl_graphs
*All PPL values are computed with the parameters `-ctk f16 -c 512 -b 4096 -ub 4096`. Changing any of these parameters will alter the PPL. In particular, reducing `-b 4096 -ub 4096` increases the PPL, while increasing them decreases the PPL.*
---
## 🚀 How do I get started?
Check out the [GGUF Tool Suite README](https://github.com/Thireus/GGUF-Tool-Suite) — focus on these sections:
1. ⚠️ **Requirements** – Which `ik_llama.cpp` (or `llama.cpp`) version to use and how to compile.
- Windows binaries (no patching needed) at: https://github.com/Thireus/ik_llama.cpp/releases
2. 📥 **Download Model Shards** – Use `quant_downloader.sh` to fetch GGUF shards from any recipe.
- Recipe examples: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
3. 🧠 **Run a Downloaded Model** – Sample usage with `llama-cli`.
4. 🛠️ **Generate a Custom Recipe** – Produce recipes tailored to your rig for optimal perplexity.
---
## ✅ Supported Models
Supported models are listed under `models/` in the [Tool Suite Github repo](https://github.com/Thireus/GGUF-Tool-Suite/tree/main/models). Presence of `ppl_results.csv` indicates official support and compatibility with `quant_assign.py`.
---
## 🤷♂️ Will I release pre-cooked GGUF files?
No, because I believe in **tailored quantization** for each user’s hardware. If you prefer ready-made shards, you are welcome to merge them via `llama-gguf-split --merge`, or request someone to publish them.
Instead, I prefer to share examples of recipes so users can see exactly how they were produced (command included inside these recipe files) and tweak them for their own rigs. The `quant_downloader.sh` script handles automatic fetching and verification of each shard. Recipes provided by [Ubergarm](https://huggingface.co/ubergarm) on his model cards are also compatible with `quant_downloader.sh`.
Users who don’t trust the GGUF shards on HuggingFace can also quantize their own by passing recipe lines to `llama-quantize --custom-q` ([see example](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/models/DeepSeek-R1-0528/DeepSeek-R1-0528-THIREUS-ANY-SPECIAL.sh#L482-L486)). Run `llama-quantize --help` to list compatible quants for `quant_assign.py`. This approach is especially useful if you prefer `llama.cpp` over `ik_llama.cpp`.
---
## 📦 What’s in this repository?
- **00001 GGUF header shard** – Contains metadata (tokens, chat template, tensor count, etc.). This metadata can be explored directly from the HuggingFace web interface after clicking on that shard.
- **Tensor shards** – Each shard holds one tensor; see `tensors.map` for names, quant types, sizes, SHA-256 hash, shard IDs, etc.
- **GPG-signed files** – `tensors.map` and header shard are signed with the key in [trusted-keys.asc](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/trusted-keys.asc) for tamper detection.
- **Security note** – Some papers about various ways to attack GGUFs and LLMs are available online, such as https://arxiv.org/abs/2505.23786, and there are also more classic security exploits like CVE-2024-23496 and CVE-2024-25664 through CVE-2024-25668. Only use GGUFs from reputable, trusted authors—or alternatively self-quantize—to avoid potential exploits.
---
## 💡 Pro Tips
You can download the BF16 model version to quantize your own shards:
```
mkdir kitchen
echo '.*=bf16' > kitchen/bf16.recipe
cd kitchen
../quant_downloader.sh bf16.recipe
```
Enjoy optimized quantization! 🎉
|
Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-IQ2_BN-SPECIAL_SPLIT
|
Thireus
| 2025-09-05T22:01:50Z | 2 | 0 | null |
[
"gguf",
"arxiv:2505.23786",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-07-24T08:14:35Z |
---
license: mit
---
# DeepSeek-TNG-R1T2-Chimera
## 🤔 What is this [HuggingFace repository](https://huggingface.co/Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_SPLIT/) about?
This repository provides **GGUF-quantized tensors** for the DeepSeek-TNG-R1T2-Chimera model (official repo: https://huggingface.co/tngtech/DeepSeek-TNG-R1T2-Chimera). These GGUF shards are designed to be used with **Thireus’ GGUF Tool Suite** (https://gguf.thireus.com), a collection of tools that automatically finds the perplexity-optimal mix of quantizations for any given VRAM and RAM target. With the Tool Suite, you can generate and download custom quantization “recipes” effortlessly.
- 📖 Read more: https://github.com/Thireus/GGUF-Tool-Suite
- 🔍 Example quant mixes: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
- 🛠️ Create your own recipe: https://colab.research.google.com/github/Thireus/GGUF-Tool-Suite/blob/main/quant_recipe_pipeline.ipynb
- 📂 Browse available quant shards: https://huggingface.co/Thireus/collections
*tl;dr: Expand the details section below*
<details>
```
cd ~
# Make sure to install all ik_llama.cpp compilation dependencies...
apt install python3-dev python3-pip python3-venv python3-wheel python3-setuptools git acl netcat-openbsd cmake # pipx
# Obtain ik_llama's Thireus version - Windows builds available at https://github.com/Thireus/ik_llama.cpp/releases
git clone https://github.com/Thireus/ik_llama.cpp
cd ik_llama.cpp
git pull
# Build ik_llama.cpp
cmake -B build -DGGML_AVX=ON -DGGML_AVX2=ON -DLLAMA_CURL=OFF -DGGML_MAX_CONTEXTS=2048
cmake --build build --config Release -j16
cd ..
# Obtain Thireus' GGUF-Tool-Suite
git clone https://github.com/Thireus/GGUF-Tool-Suite
# Download model quant mix from recipe file:
cd GGUF-Tool-Suite
rm -f download.conf # Make sure to copy the relevant download.conf for the model before running quant_assign.py
cp -f models/DeepSeek-TNG-R1T2-Chimera/download.conf . # Use the download.conf of the chosen model
mkdir -p kitchen && cd kitchen
../quant_downloader.sh ../recipe_examples/ik_llama.cpp_recipes/DeepSeek-TNG-R1T2-Chimera.ROOT-3.0624bpw-3.3657ppl.238GB-GGUF_11GB-GPU_227GB-CPU.13549e6_1ac857a.recipe
# Launch ik_llama's llama-cli:
ulimit -n 9999 # Lifts "too many open files" limitation on Linux
~/ik_llama.cpp/build/bin/llama-cli \
-m DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_TENSOR-00001-of-01148.gguf \
-mla 3 -fa -amb 512 -fmoe -ctk f16 -c 4096 -ngl 99 \
-ot "blk\.(3|4|5|6)\.ffn_.*=CUDA0" \
-ot "blk\.(7|8|9|10)\.ffn_.*=CUDA1" \
-ot exps=CPU -b 2048 -ub 1024 --warmup-batch --no-mmap --threads 36 \
--main-gpu 0 \
-p '<｜begin▁of▁sentence｜><｜User｜>What is the solution of x+5=-2?<｜Assistant｜><think>\n'
```
</details>
---
## ❓ Why does this Tool Suite exist?
1. **Compatibility & Speed** – [unsloth](https://huggingface.co/unsloth)’s dynamic quants may not always work optimally with `ik_llama.cpp`.
2. **Custom Rig Fit** – No off-the-shelf GGUF model perfectly matched my VRAM/RAM setup, so I built a way to tailor models and leverage extra VRAM/RAM to reduce perplexity.
3. **Automated PPL-Optimal Quantization** – To my knowledge, there was no flexible, automated method to minimize perplexity for any bits-per-weight (bpw) target—so I created one with excellent results!
---
## 📊 How does it compare to other GGUFs?
Here’s how DeepSeek-R1-0528 quantized with **Thireus’ GGUF Tool Suite** stacks up against other quantizers (lower perplexity = better at equal or lower bpw):
> _Note: The `recipe_examples` files illustrate good recipes. The Tool Suite computes the optimal ppl/bpw curve for you — just specify your target RAM, VRAM, and quant types, and `quant_assign.py` finds the best mix._
More perplexity/bpw graphs for other supported models: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/ppl_graphs
*All PPL values are computed with the parameters `-ctk f16 -c 512 -b 4096 -ub 4096`. Changing any of these parameters will alter the PPL. In particular, reducing `-b 4096 -ub 4096` increases the PPL, while increasing them decreases the PPL.*
---
## 🚀 How do I get started?
Check out the [GGUF Tool Suite README](https://github.com/Thireus/GGUF-Tool-Suite) — focus on these sections:
1. ⚠️ **Requirements** – Which `ik_llama.cpp` (or `llama.cpp`) version to use and how to compile.
- Windows binaries (no patching needed) at: https://github.com/Thireus/ik_llama.cpp/releases
2. 📥 **Download Model Shards** – Use `quant_downloader.sh` to fetch GGUF shards from any recipe.
- Recipe examples: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
3. 🧠 **Run a Downloaded Model** – Sample usage with `llama-cli`.
4. 🛠️ **Generate a Custom Recipe** – Produce recipes tailored to your rig for optimal perplexity.
---
## ✅ Supported Models
Supported models are listed under `models/` in the [Tool Suite Github repo](https://github.com/Thireus/GGUF-Tool-Suite/tree/main/models). Presence of `ppl_results.csv` indicates official support and compatibility with `quant_assign.py`.
---
## 🤷♂️ Will I release pre-cooked GGUF files?
No, because I believe in **tailored quantization** for each user’s hardware. If you prefer ready-made shards, you are welcome to merge them via `llama-gguf-split --merge`, or request someone to publish them.
Instead, I prefer to share examples of recipes so users can see exactly how they were produced (command included inside these recipe files) and tweak them for their own rigs. The `quant_downloader.sh` script handles automatic fetching and verification of each shard. Recipes provided by [Ubergarm](https://huggingface.co/ubergarm) on his model cards are also compatible with `quant_downloader.sh`.
Users who don’t trust the GGUF shards on HuggingFace can also quantize their own by passing recipe lines to `llama-quantize --custom-q` ([see example](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/models/DeepSeek-R1-0528/DeepSeek-R1-0528-THIREUS-ANY-SPECIAL.sh#L482-L486)). Run `llama-quantize --help` to list compatible quants for `quant_assign.py`. This approach is especially useful if you prefer `llama.cpp` over `ik_llama.cpp`.
---
## 📦 What’s in this repository?
- **00001 GGUF header shard** – Contains metadata (tokens, chat template, tensor count, etc.). This metadata can be explored directly from the HuggingFace web interface after clicking on that shard.
- **Tensor shards** – Each shard holds one tensor; see `tensors.map` for names, quant types, sizes, SHA-256 hash, shard IDs, etc.
- **GPG-signed files** – `tensors.map` and header shard are signed with the key in [trusted-keys.asc](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/trusted-keys.asc) for tamper detection.
- **Security note** – Some papers about various ways to attack GGUFs and LLMs are available online, such as https://arxiv.org/abs/2505.23786, and there are also more classic security exploits like CVE-2024-23496 and CVE-2024-25664 through CVE-2024-25668. Only use GGUFs from reputable, trusted authors—or alternatively self-quantize—to avoid potential exploits.
---
## 💡 Pro Tips
You can download the BF16 model version to quantize your own shards:
```
mkdir kitchen
echo '.*=bf16' > kitchen/bf16.recipe
cd kitchen
../quant_downloader.sh bf16.recipe
```
Enjoy optimized quantization! 🎉
|
mistpist/blockassist-bc-voracious_deadly_chameleon_1757109644
|
mistpist
| 2025-09-05T22:01:47Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"voracious deadly chameleon",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T22:01:14Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- voracious deadly chameleon
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
klmdr22/blockassist-bc-wild_loud_newt_1757109641
|
klmdr22
| 2025-09-05T22:01:23Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"wild loud newt",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T22:01:19Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- wild loud newt
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
aliangdw/rfm_prefprog_v2
|
aliangdw
| 2025-09-05T22:01:11Z | 0 | 0 |
transformers
|
[
"transformers",
"safetensors",
"qwen2_5_vl",
"reward-model",
"rfm",
"vision-language",
"multimodal",
"base_model:Qwen/Qwen2.5-VL-3B-Instruct",
"base_model:finetune:Qwen/Qwen2.5-VL-3B-Instruct",
"license:apache-2.0",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | null | 2025-09-05T21:54:53Z |
---
license: apache-2.0
base_model: Qwen/Qwen2.5-VL-3B-Instruct
tags:
- reward-model
- rfm
- vision-language
- multimodal
library_name: transformers
---
# aliangdw/rfm_prefprog_v2
This is a Reward Function Model (RFM) for vision-language preference learning and similarity assessment.
## Model Details
- **Base Model**: Qwen/Qwen2.5-VL-3B-Instruct
- **Model Type**: qwen2_5_vl
- **Architecture**: RFMModel
- **Task**: Vision-Language Reward Modeling
- **Training Method**: FSDP (Fully Sharded Data Parallel)
## Usage
```python
from transformers import AutoProcessor, AutoModel
import torch
# Load model and processor
processor = AutoProcessor.from_pretrained("aliangdw/rfm_prefprog_v2", trust_remote_code=True)
model = AutoModel.from_pretrained("aliangdw/rfm_prefprog_v2", trust_remote_code=True)
# Example usage for preference scoring
# inputs = processor(images=images, text=text, return_tensors="pt")
# outputs = model(**inputs, sample_type="preference")
```
## Model Capabilities
This RFM model can perform:
1. **Preference Prediction**: Given two trajectories A and B, predict which one is preferred
2. **Similarity Assessment**: Evaluate how similar a trajectory is to a reference
3. **Progress Estimation**: Estimate task completion progress
## Training
The model was trained using:
- FSDP for distributed training
- Mixed precision (bfloat16)
- Custom loss functions for preference and similarity learning
## Files
This repository contains:
- Model weights in SafeTensors format
- Configuration files
- Tokenizer/Processor files
## Citation
If you use this model, please cite:
|
gopterwegop/blockassist-bc-omnivorous_whiskered_skunk_1757109625
|
gopterwegop
| 2025-09-05T22:00:46Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"omnivorous whiskered skunk",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T22:00:25Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- omnivorous whiskered skunk
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
auto-space/distrostore
|
auto-space
| 2025-09-05T22:00:19Z | 0 | 0 | null |
[
"region:us"
] | null | 2025-01-02T16:01:40Z |
---
title: Distrostore
emoji: 🏢
colorFrom: green
colorTo: blue
sdk: docker
pinned: false
---
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
|
WijewardhanaNT/xnli_en_1000_2
|
WijewardhanaNT
| 2025-09-05T22:00:12Z | 0 | 0 |
transformers
|
[
"transformers",
"safetensors",
"arxiv:1910.09700",
"endpoints_compatible",
"region:us"
] | null | 2025-09-05T01:36:03Z |
---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed]
|
hogensynoo/blockassist-bc-dappled_leaping_anaconda_1757109517
|
hogensynoo
| 2025-09-05T21:58:59Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"dappled leaping anaconda",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:58:37Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- dappled leaping anaconda
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-IQ1_KT-SPECIAL_SPLIT
|
Thireus
| 2025-09-05T21:57:05Z | 1 | 0 | null |
[
"gguf",
"arxiv:2505.23786",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-07-21T06:06:55Z |
---
license: mit
---
# DeepSeek-TNG-R1T2-Chimera
## 🤔 What is this [HuggingFace repository](https://huggingface.co/Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_SPLIT/) about?
This repository provides **GGUF-quantized tensors** for the DeepSeek-TNG-R1T2-Chimera model (official repo: https://huggingface.co/tngtech/DeepSeek-TNG-R1T2-Chimera). These GGUF shards are designed to be used with **Thireus’ GGUF Tool Suite** (https://gguf.thireus.com), a collection of tools that automatically finds the perplexity-optimal mix of quantizations for any given VRAM and RAM target. With the Tool Suite, you can generate and download custom quantization “recipes” effortlessly.
- 📖 Read more: https://github.com/Thireus/GGUF-Tool-Suite
- 🔍 Example quant mixes: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
- 🛠️ Create your own recipe: https://colab.research.google.com/github/Thireus/GGUF-Tool-Suite/blob/main/quant_recipe_pipeline.ipynb
- 📂 Browse available quant shards: https://huggingface.co/Thireus/collections
*tl;dr: Expand the details section below*
<details>
```
cd ~
# Make sure to install all ik_llama.cpp compilation dependencies...
apt install python3-dev python3-pip python3-venv python3-wheel python3-setuptools git acl netcat-openbsd cmake # pipx
# Obtain ik_llama's Thireus version - Windows builds available at https://github.com/Thireus/ik_llama.cpp/releases
git clone https://github.com/Thireus/ik_llama.cpp
cd ik_llama.cpp
git pull
# Build ik_llama.cpp
cmake -B build -DGGML_AVX=ON -DGGML_AVX2=ON -DLLAMA_CURL=OFF -DGGML_MAX_CONTEXTS=2048
cmake --build build --config Release -j16
cd ..
# Obtain Thireus' GGUF-Tool-Suite
git clone https://github.com/Thireus/GGUF-Tool-Suite
# Download model quant mix from recipe file:
cd GGUF-Tool-Suite
rm -f download.conf # Make sure to copy the relevant download.conf for the model before running quant_assign.py
cp -f models/DeepSeek-TNG-R1T2-Chimera/download.conf . # Use the download.conf of the chosen model
mkdir -p kitchen && cd kitchen
../quant_downloader.sh ../recipe_examples/ik_llama.cpp_recipes/DeepSeek-TNG-R1T2-Chimera.ROOT-3.0624bpw-3.3657ppl.238GB-GGUF_11GB-GPU_227GB-CPU.13549e6_1ac857a.recipe
# Launch ik_llama's llama-cli:
ulimit -n 9999 # Lifts "too many open files" limitation on Linux
~/ik_llama.cpp/build/bin/llama-cli \
-m DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_TENSOR-00001-of-01148.gguf \
-mla 3 -fa -amb 512 -fmoe -ctk f16 -c 4096 -ngl 99 \
-ot "blk\.(3|4|5|6)\.ffn_.*=CUDA0" \
-ot "blk\.(7|8|9|10)\.ffn_.*=CUDA1" \
-ot exps=CPU -b 2048 -ub 1024 --warmup-batch --no-mmap --threads 36 \
--main-gpu 0 \
-p '<｜begin▁of▁sentence｜><｜User｜>What is the solution of x+5=-2?<｜Assistant｜><think>\n'
```
</details>
---
## ❓ Why does this Tool Suite exist?
1. **Compatibility & Speed** – [unsloth](https://huggingface.co/unsloth)’s dynamic quants may not always work optimally with `ik_llama.cpp`.
2. **Custom Rig Fit** – No off-the-shelf GGUF model perfectly matched my VRAM/RAM setup, so I built a way to tailor models and leverage extra VRAM/RAM to reduce perplexity.
3. **Automated PPL-Optimal Quantization** – To my knowledge, there was no flexible, automated method to minimize perplexity for any bits-per-weight (bpw) target—so I created one with excellent results!
---
## 📊 How does it compare to other GGUFs?
Here’s how DeepSeek-R1-0528 quantized with **Thireus’ GGUF Tool Suite** stacks up against other quantizers (lower perplexity = better at equal or lower bpw):
> _Note: The `recipe_examples` files illustrate good recipes. The Tool Suite computes the optimal ppl/bpw curve for you — just specify your target RAM, VRAM, and quant types, and `quant_assign.py` finds the best mix._
More perplexity/bpw graphs for other supported models: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/ppl_graphs
*All PPL values are computed with the parameters `-ctk f16 -c 512 -b 4096 -ub 4096`. Changing any of these parameters will alter the PPL. In particular, reducing `-b 4096 -ub 4096` increases the PPL, while increasing them decreases the PPL.*
---
## 🚀 How do I get started?
Check out the [GGUF Tool Suite README](https://github.com/Thireus/GGUF-Tool-Suite) — focus on these sections:
1. ⚠️ **Requirements** – Which `ik_llama.cpp` (or `llama.cpp`) version to use and how to compile.
- Windows binaries (no patching needed) at: https://github.com/Thireus/ik_llama.cpp/releases
2. 📥 **Download Model Shards** – Use `quant_downloader.sh` to fetch GGUF shards from any recipe.
- Recipe examples: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
3. 🧠 **Run a Downloaded Model** – Sample usage with `llama-cli`.
4. 🛠️ **Generate a Custom Recipe** – Produce recipes tailored to your rig for optimal perplexity.
---
## ✅ Supported Models
Supported models are listed under `models/` in the [Tool Suite Github repo](https://github.com/Thireus/GGUF-Tool-Suite/tree/main/models). Presence of `ppl_results.csv` indicates official support and compatibility with `quant_assign.py`.
---
## 🤷♂️ Will I release pre-cooked GGUF files?
No, because I believe in **tailored quantization** for each user’s hardware. If you prefer ready-made shards, you are welcome to merge them via `llama-gguf-split --merge`, or request someone to publish them.
Instead, I prefer to share examples of recipes so users can see exactly how they were produced (command included inside these recipe files) and tweak them for their own rigs. The `quant_downloader.sh` script handles automatic fetching and verification of each shard. Recipes provided by [Ubergarm](https://huggingface.co/ubergarm) on his model cards are also compatible with `quant_downloader.sh`.
Users who don’t trust the GGUF shards on HuggingFace can also quantize their own by passing recipe lines to `llama-quantize --custom-q` ([see example](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/models/DeepSeek-R1-0528/DeepSeek-R1-0528-THIREUS-ANY-SPECIAL.sh#L482-L486)). Run `llama-quantize --help` to list compatible quants for `quant_assign.py`. This approach is especially useful if you prefer `llama.cpp` over `ik_llama.cpp`.
---
## 📦 What’s in this repository?
- **00001 GGUF header shard** – Contains metadata (tokens, chat template, tensor count, etc.). This metadata can be explored directly from the HuggingFace web interface after clicking on that shard.
- **Tensor shards** – Each shard holds one tensor; see `tensors.map` for names, quant types, sizes, SHA-256 hash, shard IDs, etc.
- **GPG-signed files** – `tensors.map` and header shard are signed with the key in [trusted-keys.asc](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/trusted-keys.asc) for tamper detection.
- **Security note** – Some papers about various ways to attack GGUFs and LLMs are available online, such as https://arxiv.org/abs/2505.23786, and there are also more classic security exploits like CVE-2024-23496 and CVE-2024-25664 through CVE-2024-25668. Only use GGUFs from reputable, trusted authors—or alternatively self-quantize—to avoid potential exploits.
---
## 💡 Pro Tips
You can download the BF16 model version to quantize your own shards:
```
mkdir kitchen
echo '.*=bf16' > kitchen/bf16.recipe
cd kitchen
../quant_downloader.sh bf16.recipe
```
Enjoy optimized quantization! 🎉
|
bah63843/blockassist-bc-plump_fast_antelope_1757109335
|
bah63843
| 2025-09-05T21:56:21Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"plump fast antelope",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:56:17Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- plump fast antelope
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
lemonhat/Qwen2.5-7B-Instruct-t1_100k_v3_tag5_cleaned_hermes_replaced
|
lemonhat
| 2025-09-05T21:54:01Z | 18 | 0 |
transformers
|
[
"transformers",
"safetensors",
"qwen2",
"text-generation",
"llama-factory",
"full",
"generated_from_trainer",
"conversational",
"base_model:Qwen/Qwen2.5-7B-Instruct",
"base_model:finetune:Qwen/Qwen2.5-7B-Instruct",
"license:other",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] |
text-generation
| 2025-09-04T16:51:36Z |
---
library_name: transformers
license: other
base_model: Qwen/Qwen2.5-7B-Instruct
tags:
- llama-factory
- full
- generated_from_trainer
model-index:
- name: t1_100k_v3_tag5_cleaned_hermes_replaced
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# t1_100k_v3_tag5_cleaned_hermes_replaced
This model is a fine-tuned version of [Qwen/Qwen2.5-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct) on the t1_100k_v3_tag5_cleaned_hermes_replaced dataset.
It achieves the following results on the evaluation set:
- Loss: 0.2064
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-06
- train_batch_size: 1
- eval_batch_size: 1
- seed: 42
- distributed_type: multi-GPU
- num_devices: 4
- total_train_batch_size: 4
- total_eval_batch_size: 4
- optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: cosine
- num_epochs: 1
### Training results
| Training Loss | Epoch | Step | Validation Loss |
|:-------------:|:------:|:-----:|:---------------:|
| 0.3294 | 0.0056 | 100 | 0.3238 |
| 0.3463 | 0.0113 | 200 | 0.3105 |
| 0.3064 | 0.0169 | 300 | 0.2956 |
| 0.4076 | 0.0225 | 400 | 0.2921 |
| 0.2991 | 0.0281 | 500 | 0.2835 |
| 0.2524 | 0.0338 | 600 | 0.2830 |
| 0.2234 | 0.0394 | 700 | 0.2775 |
| 0.2588 | 0.0450 | 800 | 0.2749 |
| 0.2886 | 0.0507 | 900 | 0.2693 |
| 0.2361 | 0.0563 | 1000 | 0.2673 |
| 0.246 | 0.0619 | 1100 | 0.2655 |
| 0.261 | 0.0675 | 1200 | 0.2641 |
| 0.2702 | 0.0732 | 1300 | 0.2627 |
| 0.2425 | 0.0788 | 1400 | 0.2644 |
| 0.3085 | 0.0844 | 1500 | 0.2619 |
| 0.2666 | 0.0901 | 1600 | 0.2587 |
| 0.2684 | 0.0957 | 1700 | 0.2575 |
| 0.2592 | 0.1013 | 1800 | 0.2564 |
| 0.2628 | 0.1070 | 1900 | 0.2565 |
| 0.2675 | 0.1126 | 2000 | 0.2537 |
| 0.2367 | 0.1182 | 2100 | 0.2521 |
| 0.2568 | 0.1238 | 2200 | 0.2529 |
| 0.2126 | 0.1295 | 2300 | 0.2516 |
| 0.2264 | 0.1351 | 2400 | 0.2505 |
| 0.243 | 0.1407 | 2500 | 0.2476 |
| 0.2453 | 0.1464 | 2600 | 0.2501 |
| 0.2714 | 0.1520 | 2700 | 0.2487 |
| 0.2542 | 0.1576 | 2800 | 0.2466 |
| 0.2635 | 0.1632 | 2900 | 0.2465 |
| 0.2412 | 0.1689 | 3000 | 0.2445 |
| 0.2222 | 0.1745 | 3100 | 0.2448 |
| 0.2692 | 0.1801 | 3200 | 0.2445 |
| 0.2298 | 0.1858 | 3300 | 0.2443 |
| 0.2522 | 0.1914 | 3400 | 0.2426 |
| 0.2351 | 0.1970 | 3500 | 0.2429 |
| 0.1751 | 0.2026 | 3600 | 0.2418 |
| 0.2214 | 0.2083 | 3700 | 0.2419 |
| 0.2298 | 0.2139 | 3800 | 0.2395 |
| 0.242 | 0.2195 | 3900 | 0.2401 |
| 0.2372 | 0.2252 | 4000 | 0.2398 |
| 0.2554 | 0.2308 | 4100 | 0.2388 |
| 0.2172 | 0.2364 | 4200 | 0.2385 |
| 0.2365 | 0.2420 | 4300 | 0.2376 |
| 0.2689 | 0.2477 | 4400 | 0.2396 |
| 0.2177 | 0.2533 | 4500 | 0.2369 |
| 0.2956 | 0.2589 | 4600 | 0.2377 |
| 0.2396 | 0.2646 | 4700 | 0.2365 |
| 0.1959 | 0.2702 | 4800 | 0.2350 |
| 0.2658 | 0.2758 | 4900 | 0.2360 |
| 0.255 | 0.2815 | 5000 | 0.2343 |
| 0.2326 | 0.2871 | 5100 | 0.2342 |
| 0.2549 | 0.2927 | 5200 | 0.2334 |
| 0.2835 | 0.2983 | 5300 | 0.2331 |
| 0.2226 | 0.3040 | 5400 | 0.2315 |
| 0.2411 | 0.3096 | 5500 | 0.2328 |
| 0.2294 | 0.3152 | 5600 | 0.2335 |
| 0.2683 | 0.3209 | 5700 | 0.2345 |
| 0.2743 | 0.3265 | 5800 | 0.2331 |
| 0.2191 | 0.3321 | 5900 | 0.2315 |
| 0.2541 | 0.3377 | 6000 | 0.2309 |
| 0.1916 | 0.3434 | 6100 | 0.2314 |
| 0.2218 | 0.3490 | 6200 | 0.2307 |
| 0.203 | 0.3546 | 6300 | 0.2299 |
| 0.2385 | 0.3603 | 6400 | 0.2309 |
| 0.2236 | 0.3659 | 6500 | 0.2287 |
| 0.2123 | 0.3715 | 6600 | 0.2289 |
| 0.1977 | 0.3771 | 6700 | 0.2291 |
| 0.3 | 0.3828 | 6800 | 0.2281 |
| 0.2239 | 0.3884 | 6900 | 0.2284 |
| 0.219 | 0.3940 | 7000 | 0.2267 |
| 0.2036 | 0.3997 | 7100 | 0.2264 |
| 0.1947 | 0.4053 | 7200 | 0.2264 |
| 0.2035 | 0.4109 | 7300 | 0.2260 |
| 0.2443 | 0.4165 | 7400 | 0.2257 |
| 0.2316 | 0.4222 | 7500 | 0.2254 |
| 0.202 | 0.4278 | 7600 | 0.2239 |
| 0.2256 | 0.4334 | 7700 | 0.2249 |
| 0.2644 | 0.4391 | 7800 | 0.2252 |
| 0.322 | 0.4447 | 7900 | 0.2244 |
| 0.2385 | 0.4503 | 8000 | 0.2232 |
| 0.1674 | 0.4560 | 8100 | 0.2236 |
| 0.2607 | 0.4616 | 8200 | 0.2229 |
| 0.2071 | 0.4672 | 8300 | 0.2232 |
| 0.2537 | 0.4728 | 8400 | 0.2216 |
| 0.2196 | 0.4785 | 8500 | 0.2213 |
| 0.21 | 0.4841 | 8600 | 0.2218 |
| 0.3098 | 0.4897 | 8700 | 0.2214 |
| 0.2339 | 0.4954 | 8800 | 0.2201 |
| 0.2187 | 0.5010 | 8900 | 0.2199 |
| 0.2026 | 0.5066 | 9000 | 0.2196 |
| 0.2132 | 0.5122 | 9100 | 0.2192 |
| 0.2218 | 0.5179 | 9200 | 0.2201 |
| 0.2152 | 0.5235 | 9300 | 0.2185 |
| 0.1799 | 0.5291 | 9400 | 0.2192 |
| 0.2413 | 0.5348 | 9500 | 0.2188 |
| 0.2345 | 0.5404 | 9600 | 0.2178 |
| 0.2336 | 0.5460 | 9700 | 0.2175 |
| 0.1982 | 0.5516 | 9800 | 0.2169 |
| 0.235 | 0.5573 | 9900 | 0.2175 |
| 0.2195 | 0.5629 | 10000 | 0.2173 |
| 0.2137 | 0.5685 | 10100 | 0.2168 |
| 0.2 | 0.5742 | 10200 | 0.2163 |
| 0.3196 | 0.5798 | 10300 | 0.2167 |
| 0.2799 | 0.5854 | 10400 | 0.2166 |
| 0.2432 | 0.5910 | 10500 | 0.2164 |
| 0.2329 | 0.5967 | 10600 | 0.2156 |
| 0.2518 | 0.6023 | 10700 | 0.2157 |
| 0.2601 | 0.6079 | 10800 | 0.2154 |
| 0.2103 | 0.6136 | 10900 | 0.2151 |
| 0.1983 | 0.6192 | 11000 | 0.2153 |
| 0.2313 | 0.6248 | 11100 | 0.2141 |
| 0.1924 | 0.6305 | 11200 | 0.2145 |
| 0.212 | 0.6361 | 11300 | 0.2143 |
| 0.2122 | 0.6417 | 11400 | 0.2142 |
| 0.2781 | 0.6473 | 11500 | 0.2136 |
| 0.2388 | 0.6530 | 11600 | 0.2140 |
| 0.2366 | 0.6586 | 11700 | 0.2132 |
| 0.2267 | 0.6642 | 11800 | 0.2130 |
| 0.2228 | 0.6699 | 11900 | 0.2123 |
| 0.1946 | 0.6755 | 12000 | 0.2117 |
| 0.2098 | 0.6811 | 12100 | 0.2119 |
| 0.1994 | 0.6867 | 12200 | 0.2120 |
| 0.1836 | 0.6924 | 12300 | 0.2119 |
| 0.2249 | 0.6980 | 12400 | 0.2114 |
| 0.1974 | 0.7036 | 12500 | 0.2114 |
| 0.26 | 0.7093 | 12600 | 0.2112 |
| 0.1836 | 0.7149 | 12700 | 0.2107 |
| 0.2052 | 0.7205 | 12800 | 0.2107 |
| 0.1848 | 0.7261 | 12900 | 0.2098 |
| 0.232 | 0.7318 | 13000 | 0.2101 |
| 0.2363 | 0.7374 | 13100 | 0.2099 |
| 0.2244 | 0.7430 | 13200 | 0.2097 |
| 0.2046 | 0.7487 | 13300 | 0.2095 |
| 0.1782 | 0.7543 | 13400 | 0.2096 |
| 0.1824 | 0.7599 | 13500 | 0.2097 |
| 0.1678 | 0.7656 | 13600 | 0.2093 |
| 0.2104 | 0.7712 | 13700 | 0.2091 |
| 0.2023 | 0.7768 | 13800 | 0.2086 |
| 0.2202 | 0.7824 | 13900 | 0.2085 |
| 0.2481 | 0.7881 | 14000 | 0.2082 |
| 0.223 | 0.7937 | 14100 | 0.2084 |
| 0.2575 | 0.7993 | 14200 | 0.2082 |
| 0.1704 | 0.8050 | 14300 | 0.2081 |
| 0.2602 | 0.8106 | 14400 | 0.2080 |
| 0.1833 | 0.8162 | 14500 | 0.2082 |
| 0.2317 | 0.8218 | 14600 | 0.2078 |
| 0.1921 | 0.8275 | 14700 | 0.2077 |
| 0.2226 | 0.8331 | 14800 | 0.2075 |
| 0.2023 | 0.8387 | 14900 | 0.2074 |
| 0.2457 | 0.8444 | 15000 | 0.2073 |
| 0.1907 | 0.8500 | 15100 | 0.2071 |
| 0.239 | 0.8556 | 15200 | 0.2072 |
| 0.2125 | 0.8612 | 15300 | 0.2071 |
| 0.2136 | 0.8669 | 15400 | 0.2070 |
| 0.1933 | 0.8725 | 15500 | 0.2069 |
| 0.2189 | 0.8781 | 15600 | 0.2069 |
| 0.2317 | 0.8838 | 15700 | 0.2068 |
| 0.187 | 0.8894 | 15800 | 0.2067 |
| 0.1828 | 0.8950 | 15900 | 0.2067 |
| 0.1873 | 0.9006 | 16000 | 0.2067 |
| 0.1995 | 0.9063 | 16100 | 0.2066 |
| 0.1763 | 0.9119 | 16200 | 0.2066 |
| 0.1942 | 0.9175 | 16300 | 0.2065 |
| 0.1666 | 0.9232 | 16400 | 0.2065 |
| 0.2616 | 0.9288 | 16500 | 0.2065 |
| 0.1909 | 0.9344 | 16600 | 0.2065 |
| 0.1878 | 0.9401 | 16700 | 0.2064 |
| 0.1995 | 0.9457 | 16800 | 0.2065 |
| 0.1973 | 0.9513 | 16900 | 0.2064 |
| 0.1855 | 0.9569 | 17000 | 0.2063 |
| 0.2068 | 0.9626 | 17100 | 0.2064 |
| 0.2285 | 0.9682 | 17200 | 0.2063 |
| 0.2533 | 0.9738 | 17300 | 0.2064 |
| 0.224 | 0.9795 | 17400 | 0.2063 |
| 0.2149 | 0.9851 | 17500 | 0.2064 |
| 0.2333 | 0.9907 | 17600 | 0.2064 |
| 0.2123 | 0.9963 | 17700 | 0.2064 |
### Framework versions
- Transformers 4.46.1
- Pytorch 2.6.0+cu124
- Datasets 3.1.0
- Tokenizers 0.20.3
|
Nouserenabel/my-sentiment-model
|
Nouserenabel
| 2025-09-05T21:53:22Z | 0 | 0 |
transformers
|
[
"transformers",
"tensorboard",
"safetensors",
"distilbert",
"text-classification",
"generated_from_trainer",
"base_model:distilbert/distilbert-base-uncased",
"base_model:finetune:distilbert/distilbert-base-uncased",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] |
text-classification
| 2025-09-05T21:26:52Z |
---
library_name: transformers
license: apache-2.0
base_model: distilbert-base-uncased
tags:
- generated_from_trainer
model-index:
- name: my-sentiment-model
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# my-sentiment-model
This model is a fine-tuned version of [distilbert-base-uncased](https://huggingface.co/distilbert-base-uncased) on an unknown dataset.
It achieves the following results on the evaluation set:
- eval_loss: 0.4832
- eval_accuracy: 0.8888
- eval_runtime: 12.2004
- eval_samples_per_second: 71.473
- eval_steps_per_second: 4.508
- epoch: 0.0523
- step: 220
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- num_epochs: 2
### Framework versions
- Transformers 4.56.1
- Pytorch 2.8.0+cu126
- Datasets 4.0.0
- Tokenizers 0.22.0
|
NahedDom/blockassist-bc-flapping_stocky_leopard_1757106860
|
NahedDom
| 2025-09-05T21:53:07Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"flapping stocky leopard",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:53:02Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- flapping stocky leopard
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
Stasonelison/blockassist-bc-howling_powerful_aardvark_1757109144
|
Stasonelison
| 2025-09-05T21:53:05Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"howling powerful aardvark",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:52:56Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- howling powerful aardvark
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-Q6_0_R4-SPECIAL_SPLIT
|
Thireus
| 2025-09-05T21:50:59Z | 1 | 0 | null |
[
"gguf",
"arxiv:2505.23786",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-07-22T13:51:13Z |
---
license: mit
---
# DeepSeek-TNG-R1T2-Chimera
## 🤔 What is this [HuggingFace repository](https://huggingface.co/Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_SPLIT/) about?
This repository provides **GGUF-quantized tensors** for the DeepSeek-TNG-R1T2-Chimera model (official repo: https://huggingface.co/tngtech/DeepSeek-TNG-R1T2-Chimera). These GGUF shards are designed to be used with **Thireus’ GGUF Tool Suite** (https://gguf.thireus.com), a collection of tools that automatically finds the perplexity-optimal mix of quantizations for any given VRAM and RAM target. With the Tool Suite, you can generate and download custom quantization “recipes” effortlessly.
- 📖 Read more: https://github.com/Thireus/GGUF-Tool-Suite
- 🔍 Example quant mixes: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
- 🛠️ Create your own recipe: https://colab.research.google.com/github/Thireus/GGUF-Tool-Suite/blob/main/quant_recipe_pipeline.ipynb
- 📂 Browse available quant shards: https://huggingface.co/Thireus/collections
*tl;dr: Expand the details section below*
<details>
```
cd ~
# Make sure to install all ik_llama.cpp compilation dependencies...
apt install python3-dev python3-pip python3-venv python3-wheel python3-setuptools git acl netcat-openbsd cmake # pipx
# Obtain ik_llama's Thireus version - Windows builds available at https://github.com/Thireus/ik_llama.cpp/releases
git clone https://github.com/Thireus/ik_llama.cpp
cd ik_llama.cpp
git pull
# Build ik_llama.cpp
cmake -B build -DGGML_AVX=ON -DGGML_AVX2=ON -DLLAMA_CURL=OFF -DGGML_MAX_CONTEXTS=2048
cmake --build build --config Release -j16
cd ..
# Obtain Thireus' GGUF-Tool-Suite
git clone https://github.com/Thireus/GGUF-Tool-Suite
# Download model quant mix from recipe file:
cd GGUF-Tool-Suite
rm -f download.conf # Make sure to copy the relevant download.conf for the model before running quant_assign.py
cp -f models/DeepSeek-TNG-R1T2-Chimera/download.conf . # Use the download.conf of the chosen model
mkdir -p kitchen && cd kitchen
../quant_downloader.sh ../recipe_examples/ik_llama.cpp_recipes/DeepSeek-TNG-R1T2-Chimera.ROOT-3.0624bpw-3.3657ppl.238GB-GGUF_11GB-GPU_227GB-CPU.13549e6_1ac857a.recipe
# Launch ik_llama's llama-cli:
ulimit -n 9999 # Lifts "too many open files" limitation on Linux
~/ik_llama.cpp/build/bin/llama-cli \
-m DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_TENSOR-00001-of-01148.gguf \
-mla 3 -fa -amb 512 -fmoe -ctk f16 -c 4096 -ngl 99 \
-ot "blk\.(3|4|5|6)\.ffn_.*=CUDA0" \
-ot "blk\.(7|8|9|10)\.ffn_.*=CUDA1" \
-ot exps=CPU -b 2048 -ub 1024 --warmup-batch --no-mmap --threads 36 \
--main-gpu 0 \
-p '<｜begin▁of▁sentence｜><｜User｜>What is the solution of x+5=-2?<｜Assistant｜><think>\n'
```
</details>
---
## ❓ Why does this Tool Suite exist?
1. **Compatibility & Speed** – [unsloth](https://huggingface.co/unsloth)’s dynamic quants may not always work optimally with `ik_llama.cpp`.
2. **Custom Rig Fit** – No off-the-shelf GGUF model perfectly matched my VRAM/RAM setup, so I built a way to tailor models and leverage extra VRAM/RAM to reduce perplexity.
3. **Automated PPL-Optimal Quantization** – To my knowledge, there was no flexible, automated method to minimize perplexity for any bits-per-weight (bpw) target—so I created one with excellent results!
---
## 📊 How does it compare to other GGUFs?
Here’s how DeepSeek-R1-0528 quantized with **Thireus’ GGUF Tool Suite** stacks up against other quantizers (lower perplexity = better at equal or lower bpw):
> _Note: The `recipe_examples` files illustrate good recipes. The Tool Suite computes the optimal ppl/bpw curve for you — just specify your target RAM, VRAM, and quant types, and `quant_assign.py` finds the best mix._
More perplexity/bpw graphs for other supported models: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/ppl_graphs
*All PPL values are computed with the parameters `-ctk f16 -c 512 -b 4096 -ub 4096`. Changing any of these parameters will alter the PPL. In particular, reducing `-b 4096 -ub 4096` increases the PPL, while increasing them decreases the PPL.*
---
## 🚀 How do I get started?
Check out the [GGUF Tool Suite README](https://github.com/Thireus/GGUF-Tool-Suite) — focus on these sections:
1. ⚠️ **Requirements** – Which `ik_llama.cpp` (or `llama.cpp`) version to use and how to compile.
- Windows binaries (no patching needed) at: https://github.com/Thireus/ik_llama.cpp/releases
2. 📥 **Download Model Shards** – Use `quant_downloader.sh` to fetch GGUF shards from any recipe.
- Recipe examples: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
3. 🧠 **Run a Downloaded Model** – Sample usage with `llama-cli`.
4. 🛠️ **Generate a Custom Recipe** – Produce recipes tailored to your rig for optimal perplexity.
---
## ✅ Supported Models
Supported models are listed under `models/` in the [Tool Suite Github repo](https://github.com/Thireus/GGUF-Tool-Suite/tree/main/models). Presence of `ppl_results.csv` indicates official support and compatibility with `quant_assign.py`.
---
## 🤷♂️ Will I release pre-cooked GGUF files?
No, because I believe in **tailored quantization** for each user’s hardware. If you prefer ready-made shards, you are welcome to merge them via `llama-gguf-split --merge`, or request someone to publish them.
Instead, I prefer to share examples of recipes so users can see exactly how they were produced (command included inside these recipe files) and tweak them for their own rigs. The `quant_downloader.sh` script handles automatic fetching and verification of each shard. Recipes provided by [Ubergarm](https://huggingface.co/ubergarm) on his model cards are also compatible with `quant_downloader.sh`.
Users who don’t trust the GGUF shards on HuggingFace can also quantize their own by passing recipe lines to `llama-quantize --custom-q` ([see example](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/models/DeepSeek-R1-0528/DeepSeek-R1-0528-THIREUS-ANY-SPECIAL.sh#L482-L486)). Run `llama-quantize --help` to list compatible quants for `quant_assign.py`. This approach is especially useful if you prefer `llama.cpp` over `ik_llama.cpp`.
---
## 📦 What’s in this repository?
- **00001 GGUF header shard** – Contains metadata (tokens, chat template, tensor count, etc.). This metadata can be explored directly from the HuggingFace web interface after clicking on that shard.
- **Tensor shards** – Each shard holds one tensor; see `tensors.map` for names, quant types, sizes, SHA-256 hash, shard IDs, etc.
- **GPG-signed files** – `tensors.map` and header shard are signed with the key in [trusted-keys.asc](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/trusted-keys.asc) for tamper detection.
- **Security note** – Some papers about various ways to attack GGUFs and LLMs are available online, such as https://arxiv.org/abs/2505.23786, and there are also more classic security exploits like CVE-2024-23496 and CVE-2024-25664 through CVE-2024-25668. Only use GGUFs from reputable, trusted authors—or alternatively self-quantize—to avoid potential exploits.
---
## 💡 Pro Tips
You can download the BF16 model version to quantize your own shards:
```
mkdir kitchen
echo '.*=bf16' > kitchen/bf16.recipe
cd kitchen
../quant_downloader.sh bf16.recipe
```
Enjoy optimized quantization! 🎉
|
ucf-crcv/GAEA-7B
|
ucf-crcv
| 2025-09-05T21:50:52Z | 0 | 3 | null |
[
"dataset:ucf-crcv/GAEA-Train",
"arxiv:2503.16423",
"base_model:Qwen/Qwen2.5-VL-7B-Instruct",
"base_model:finetune:Qwen/Qwen2.5-VL-7B-Instruct",
"license:cc",
"region:us"
] | null | 2025-03-15T05:48:51Z |
---
license: cc
datasets:
- ucf-crcv/GAEA-Train
base_model:
- Qwen/Qwen2.5-VL-7B-Instruct
---
<h1 align="left"> GAEA: A Geolocation Aware Conversational Assistant [WACV 2026🔥]</h1>
<h3 align="left"> Summary</h3>
<p align="justify"> Image geolocalization, in which an AI model traditionally predicts the precise GPS coordinates of an image, is a challenging task with many downstream applications. However, the user cannot utilize the model to further their knowledge beyond the GPS coordinates; the model lacks an understanding of the location and the conversational ability to communicate with the user. In recent days, with the tremendous progress of large multimodal models (LMMs) — proprietary and open-source — researchers have attempted to geolocalize images via LMMs. However, the issues remain unaddressed; beyond general tasks, for more specialized downstream tasks, such as geolocalization, LMMs struggle. In this work, we propose solving this problem by introducing a conversational model, GAEA, that provides information regarding the location of an image as the user requires. No large-scale dataset enabling the training of such a model exists. Thus, we propose GAEA-1.4M, a comprehensive dataset comprising over 800k images and approximately 1.4M question-answer pairs, constructed by leveraging OpenStreetMap (OSM) attributes and geographical context clues. For quantitative evaluation, we propose a diverse benchmark, GAEA-Bench, comprising 3.5k image-text pairs to evaluate conversational capabilities equipped with diverse question types. We consider 11 state-of-the-art open-source and proprietary LMMs and demonstrate that GAEA significantly outperforms the best open-source model, LLaVA-OneVision, by 18.2% and the best proprietary model, GPT-4o, by 7.2%. Our dataset, model, and codes are publicly available. </p>
## `GAEA` is the first open-source conversational model for conversational capabilities equipped with global-scale geolocalization.
[](https://arxiv.org/abs/2503.16423)
[](https://huggingface.co/collections/ucf-crcv/gaea-67d514a61d48eb1708b13a08)
[](https://ucf-crcv.github.io/GAEA/)
**Main contributions:**
1) **`GAEA-Train: A Diverse Training Dataset:`** We propose GAEA-Train, a new dataset designed for training conversational image geolocalization models, incorporating diverse visual and contextual data.
2) **`GAEA-Bench: Evaluating Conversational Geolocalization:`** To assess conversational capabilities in geolocalization, we introduce GAEA-Bench, a benchmark featuring various question-answer formats.
3) **`GAEA: An Interactive Geolocalization Chatbot:`** We present GAEA, a conversational chatbot that extends beyond geolocalization to provide rich contextual insights about locations from images.
4) **`Benchmarking Against State-of-the-Art LMMs:`** We quantitatively compare our model’s performance against 8 open-source and 3 proprietary LMMs, including GPT-4o and Gemini-2.0-Flash.
<b> This page is dedicated to the GAEA model </b>
<p align="center">
<img src="Assets/teaser.jpg" alt="teaser" width="800px"/></a>
</p>
<p align="justify"> We compare the performance of various LMMs on the geographically-grounded visual-question-answering task, included in our new GAEA-Bench benchmark. Most LMMs can describe the Wat Pho statue, but only GAEA, our Geolocation Aware Assistant, retrieves the correct nearby cafe, Cafe Amazon <i>(left)</i>. Qualitative SVQA comparison showing GAEA’s ability to provide accurate, location-specific answers where other LMMs fail <i>(right)</i>.</p>
<h2 align="left"> Model Description</h2>
<h3 align="left">Architecture</h3>
<p align="left"><img src="Assets/arch.png" alt="arch" width="400px"/></p>
<p align="justify"> <b>Overview of the GAEA model architecture and workflow.</b> An input image is first processed by a Vision Transformer (ViT) encoder, whose output is projected through a visual projector to obtain visual embeddings. Simultaneously, the input text prompt is converted into text embeddings. The combined visual and textual embeddings are then fed into the Qwen2.5 LLM space, which generates a response based on the multimodal input. We follow the single-stage training approach, unfreezing MLP, and performing LoRA fine-tuning in the same stage. </p>
<!-- <h2 align="left"> How To Use</h2> -->
<h2 align="left">Evaluation Results</h2>
<h3 align="left">Comparison with SoTA LMMs on GAEA-Bench (Conversational) </h3>
<p align="left">
<img src="Assets/GAEA-Benc-Eval.png" alt="GAEA-Benc-Eval" width="500px"/></a>
</p>
<p align="justify"> We benchmark 11 open-source and proprietary LMMs on GAEA-Bench. Notably, GAEA outperforms all open-source models and fares higher than the proprietary models on decision-making questions (MCQs and TFs). We provide the relative performance change for each model compared to GAEA. We use GPT-4o as a judge for evaluation, and it has been documented that LLMs as judges prefer their long-form output; hence, the scores for these models are likely overestimated. </p>
<p align="left">
<img src="Assets/question_types_stats.jpg" alt="question-types-stats" width="500px"/></a>
</p>
<p align="justify">We showcase the performance of various LMMs on four diverse question types. GAEA outperforms on average across all question forms.</p>
<h3 align="left">Qualitative Results (Conversational) </h3>
<p align="left">
<img src="Assets/queston_types_qual.jpg" alt="queston-types-qual" width="500px"/></a>
</p>
<p align="justify"> Qualitative MCQs comparison showing GAEA’s ability to provide accurate answers where other LMMs fail. </p>
<h3 align="left">Comparison with Specialized Models on Standard Geolocalization Datasets</h3>
<p align="left">
<img src="Assets/Geolocalization_results.png" alt="Geolocalization_results" width="400px"/></a>
</p>
<p align="justify"> We benchmark the performance of various specialized models on standard geolocation datasets. GAEA demonstrates competitive results, outperforming GaGA on multiple distance thresholds in both IM2GPS and IM2GPS3k. </p>
<h3 align="left">Comparison with best SoTA LMMs on City/Country Prediction </h3>
<p align="left">
<img src="Assets/City_Country_results.jpg" alt="City-Country-results" width="400px"/></a>
</p>
<p align="justify"> Classification accuracy for both city and country labels, where GAEA surpasses several recent LMMs in performance. </p>
---
# Citation
**BibTeX:**
```bibtex
@misc{campos2025gaeageolocationawareconversational,
title={GAEA: A Geolocation Aware Conversational Assistant},
author={Ron Campos and Ashmal Vayani and Parth Parag Kulkarni and Rohit Gupta and Aritra Dutta and Mubarak Shah},
year={2025},
eprint={2503.16423},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2503.16423},
}
```
---
## Licensing Information
We release our work under [CC BY-NC 4.0 License](https://creativecommons.org/licenses/by-nc/4.0/). The CC BY-NC 4.0 license allows others to share, remix, and adapt the work, as long as it's for non-commercial purposes and proper attribution is given to the original creator.
|
rocktanmay2012/blockassist-bc-bold_placid_barracuda_1757108968
|
rocktanmay2012
| 2025-09-05T21:50:08Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"bold placid barracuda",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:49:53Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- bold placid barracuda
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
cwayneconnor/blockassist-bc-mute_loud_lynx_1757108749
|
cwayneconnor
| 2025-09-05T21:49:47Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"mute loud lynx",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:47:08Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- mute loud lynx
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-Q6_0-SPECIAL_SPLIT
|
Thireus
| 2025-09-05T21:49:45Z | 2 | 0 | null |
[
"gguf",
"arxiv:2505.23786",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-07-23T16:19:16Z |
---
license: mit
---
# DeepSeek-TNG-R1T2-Chimera
## 🤔 What is this [HuggingFace repository](https://huggingface.co/Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_SPLIT/) about?
This repository provides **GGUF-quantized tensors** for the DeepSeek-TNG-R1T2-Chimera model (official repo: https://huggingface.co/tngtech/DeepSeek-TNG-R1T2-Chimera). These GGUF shards are designed to be used with **Thireus’ GGUF Tool Suite** (https://gguf.thireus.com), a collection of tools that automatically finds the perplexity-optimal mix of quantizations for any given VRAM and RAM target. With the Tool Suite, you can generate and download custom quantization “recipes” effortlessly.
- 📖 Read more: https://github.com/Thireus/GGUF-Tool-Suite
- 🔍 Example quant mixes: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
- 🛠️ Create your own recipe: https://colab.research.google.com/github/Thireus/GGUF-Tool-Suite/blob/main/quant_recipe_pipeline.ipynb
- 📂 Browse available quant shards: https://huggingface.co/Thireus/collections
*tl;dr: Expand the details section below*
<details>
```
cd ~
# Make sure to install all ik_llama.cpp compilation dependencies...
apt install python3-dev python3-pip python3-venv python3-wheel python3-setuptools git acl netcat-openbsd cmake # pipx
# Obtain ik_llama's Thireus version - Windows builds available at https://github.com/Thireus/ik_llama.cpp/releases
git clone https://github.com/Thireus/ik_llama.cpp
cd ik_llama.cpp
git pull
# Build ik_llama.cpp
cmake -B build -DGGML_AVX=ON -DGGML_AVX2=ON -DLLAMA_CURL=OFF -DGGML_MAX_CONTEXTS=2048
cmake --build build --config Release -j16
cd ..
# Obtain Thireus' GGUF-Tool-Suite
git clone https://github.com/Thireus/GGUF-Tool-Suite
# Download model quant mix from recipe file:
cd GGUF-Tool-Suite
rm -f download.conf # Make sure to copy the relevant download.conf for the model before running quant_assign.py
cp -f models/DeepSeek-TNG-R1T2-Chimera/download.conf . # Use the download.conf of the chosen model
mkdir -p kitchen && cd kitchen
../quant_downloader.sh ../recipe_examples/ik_llama.cpp_recipes/DeepSeek-TNG-R1T2-Chimera.ROOT-3.0624bpw-3.3657ppl.238GB-GGUF_11GB-GPU_227GB-CPU.13549e6_1ac857a.recipe
# Launch ik_llama's llama-cli:
ulimit -n 9999 # Lifts "too many open files" limitation on Linux
~/ik_llama.cpp/build/bin/llama-cli \
-m DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_TENSOR-00001-of-01148.gguf \
-mla 3 -fa -amb 512 -fmoe -ctk f16 -c 4096 -ngl 99 \
-ot "blk\.(3|4|5|6)\.ffn_.*=CUDA0" \
-ot "blk\.(7|8|9|10)\.ffn_.*=CUDA1" \
-ot exps=CPU -b 2048 -ub 1024 --warmup-batch --no-mmap --threads 36 \
--main-gpu 0 \
-p '<｜begin▁of▁sentence｜><｜User｜>What is the solution of x+5=-2?<｜Assistant｜><think>\n'
```
</details>
---
## ❓ Why does this Tool Suite exist?
1. **Compatibility & Speed** – [unsloth](https://huggingface.co/unsloth)’s dynamic quants may not always work optimally with `ik_llama.cpp`.
2. **Custom Rig Fit** – No off-the-shelf GGUF model perfectly matched my VRAM/RAM setup, so I built a way to tailor models and leverage extra VRAM/RAM to reduce perplexity.
3. **Automated PPL-Optimal Quantization** – To my knowledge, there was no flexible, automated method to minimize perplexity for any bits-per-weight (bpw) target—so I created one with excellent results!
---
## 📊 How does it compare to other GGUFs?
Here’s how DeepSeek-R1-0528 quantized with **Thireus’ GGUF Tool Suite** stacks up against other quantizers (lower perplexity = better at equal or lower bpw):
> _Note: The `recipe_examples` files illustrate good recipes. The Tool Suite computes the optimal ppl/bpw curve for you — just specify your target RAM, VRAM, and quant types, and `quant_assign.py` finds the best mix._
More perplexity/bpw graphs for other supported models: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/ppl_graphs
*All PPL values are computed with the parameters `-ctk f16 -c 512 -b 4096 -ub 4096`. Changing any of these parameters will alter the PPL. In particular, reducing `-b 4096 -ub 4096` increases the PPL, while increasing them decreases the PPL.*
---
## 🚀 How do I get started?
Check out the [GGUF Tool Suite README](https://github.com/Thireus/GGUF-Tool-Suite) — focus on these sections:
1. ⚠️ **Requirements** – Which `ik_llama.cpp` (or `llama.cpp`) version to use and how to compile.
- Windows binaries (no patching needed) at: https://github.com/Thireus/ik_llama.cpp/releases
2. 📥 **Download Model Shards** – Use `quant_downloader.sh` to fetch GGUF shards from any recipe.
- Recipe examples: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
3. 🧠 **Run a Downloaded Model** – Sample usage with `llama-cli`.
4. 🛠️ **Generate a Custom Recipe** – Produce recipes tailored to your rig for optimal perplexity.
---
## ✅ Supported Models
Supported models are listed under `models/` in the [Tool Suite Github repo](https://github.com/Thireus/GGUF-Tool-Suite/tree/main/models). Presence of `ppl_results.csv` indicates official support and compatibility with `quant_assign.py`.
---
## 🤷♂️ Will I release pre-cooked GGUF files?
No, because I believe in **tailored quantization** for each user’s hardware. If you prefer ready-made shards, you are welcome to merge them via `llama-gguf-split --merge`, or request someone to publish them.
Instead, I prefer to share examples of recipes so users can see exactly how they were produced (command included inside these recipe files) and tweak them for their own rigs. The `quant_downloader.sh` script handles automatic fetching and verification of each shard. Recipes provided by [Ubergarm](https://huggingface.co/ubergarm) on his model cards are also compatible with `quant_downloader.sh`.
Users who don’t trust the GGUF shards on HuggingFace can also quantize their own by passing recipe lines to `llama-quantize --custom-q` ([see example](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/models/DeepSeek-R1-0528/DeepSeek-R1-0528-THIREUS-ANY-SPECIAL.sh#L482-L486)). Run `llama-quantize --help` to list compatible quants for `quant_assign.py`. This approach is especially useful if you prefer `llama.cpp` over `ik_llama.cpp`.
---
## 📦 What’s in this repository?
- **00001 GGUF header shard** – Contains metadata (tokens, chat template, tensor count, etc.). This metadata can be explored directly from the HuggingFace web interface after clicking on that shard.
- **Tensor shards** – Each shard holds one tensor; see `tensors.map` for names, quant types, sizes, SHA-256 hash, shard IDs, etc.
- **GPG-signed files** – `tensors.map` and header shard are signed with the key in [trusted-keys.asc](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/trusted-keys.asc) for tamper detection.
- **Security note** – Some papers about various ways to attack GGUFs and LLMs are available online, such as https://arxiv.org/abs/2505.23786, and there are also more classic security exploits like CVE-2024-23496 and CVE-2024-25664 through CVE-2024-25668. Only use GGUFs from reputable, trusted authors—or alternatively self-quantize—to avoid potential exploits.
---
## 💡 Pro Tips
You can download the BF16 model version to quantize your own shards:
```
mkdir kitchen
echo '.*=bf16' > kitchen/bf16.recipe
cd kitchen
../quant_downloader.sh bf16.recipe
```
Enjoy optimized quantization! 🎉
|
EarthnDusk/Loras_KtiseosNyx
|
EarthnDusk
| 2025-09-05T21:49:14Z | 0 | 0 |
diffusers
|
[
"diffusers",
"text-to-image",
"dataset:EarthnDusk/XL_PDXL_Embeddings",
"base_model:OnomaAIResearch/Illustrious-xl-early-release-v0",
"base_model:finetune:OnomaAIResearch/Illustrious-xl-early-release-v0",
"license:creativeml-openrail-m",
"region:us"
] |
text-to-image
| 2025-09-05T00:19:58Z |
---
license: creativeml-openrail-m
datasets:
- EarthnDusk/XL_PDXL_Embeddings
base_model:
- OnomaAIResearch/Illustrious-xl-early-release-v0
pipeline_tag: text-to-image
library_name: diffusers
---
<style>
.custom-table td {
width: 33.333%;
}
.custom-image-container {
position: relative;
width: 100%;
height: 100%;
border-radius: 0.5em;
overflow: hidden;
align-items: center;
}
.custom-image {
width: 100%;
height: auto;
border-radius: 0.5em;
transition: transform 0.25s;
}
.custom-image-container:hover .custom-image {
transform: scale(1.2);
}
/* Style for tables within Markdown. Makes them look nicer. */
.markdown table {
border-collapse: collapse; /* Collapse borders for a cleaner look */
width: 100%; /* Take up full width */
margin-bottom: 1em; /* Add space after the table */
}
.markdown th,
.markdown td {
border: 1px solid #ddd; /* Subtle borders */
padding: 8px; /* Add padding for readability */
text-align: left; /* Left-align text */
}
.markdown th {
background-color: #f2f2f2; /* Light gray background for headers */
font-weight: bold; /* Bold header text */
}
/* Style for summary elements */
summary {
cursor: pointer;
font-weight: bold;
margin-bottom: 0.5em; /* Adds space for visual clarity */
}
</style>
# Loras
Ktiseos Nyx Loras! These aren't just backups these are ones we've been training since our 2025 repo got pretty full.
While these are free for you to download and use at your own discretion based on how open source should be...
We would adere to the fact that if you could donate money for the time it took to train these items!
To find the keywords for the lora you just use Xypher's tool here: https://xypher7.github.io/lora-metadata-viewer/
These are LARGELY for Stable Diffusion XL base - such as Illustrious & Pony XL as well as NoobAI.
# Previews
The previews in this container are not yet named, give me time, i'll sort it out lol, I am borrowing code from Holostrawberry that he uses on HolyMix!
Also some are from teh old repo, so i'm still working on bringing previews in
<table class="custom-table">
<tr>
<td>
<div class="custom-image-container">
<img class="custom-image" src="Rogue%20Lora%202025/image%20-%202025-03-18T142916.899.jpeg" alt="Preview">
</div>
</td>
<td>
<div class="custom-image-container">
<img class="custom-image" src="Rogue%20Lora%202025/image%20-%202025-03-18T143928.899.jpeg" alt="Preview">
</div>
</td>
<td>
<div class="custom-image-container">
<img class="custom-image" src="Aion%20RPG/image%20-%202025-03-15T211741.893.jpeg" alt="Preview">
</div>
</td>
</tr>
<tr>
<td>
<div class="custom-image-container">
<img class="custom-image" src="LoraPreviews/image%20-%202025-03-05T183052.609.jpeg" alt="Preview">
</div>
</td>
<td>
<div class="custom-image-container">
<img class="custom-image" src="StaticBloomStyle%20PDXL%20Samples/image%20-%202025-03-11T111302.898.jpeg" alt="Preview">
</div>
</td>
<td>
<div class="custom-image-container">
<img class="custom-image" src="Aion%20RPG/image%20-%202025-03-15T213638.429.jpeg" alt="Preview">
</div>
</td>
</tr>
<tr>
<td>
<div class="custom-image-container">
<img class="custom-image" src="LoraPreviews/image%20-%202025-03-05T204313.711.jpeg" alt="Preview">
</div>
</td>
<td>
<div class="custom-image-container">
<img class="custom-image" src="LoraPreviews/image%20-%202025-03-05T205102.527.jpeg" alt="Preview">
</div>
</td>
<td>
<div class="custom-image-container">
<img class="custom-image" src="Aion%20RPG/image%20-%202025-03-15T213727.538.jpeg" alt="Preview">
</div>
</td>
</tr>
<tr>
<td>
<div class="custom-image-container">
<img class="custom-image" src="https://huggingface.co/EarthnDusk/Loras_2025/resolve/main/Arcane%20Pony%20Samples/image%20-%202025-04-08T193422.100.jpeg" alt="Preview">
</div>
</td>
<td>
<div class="custom-image-container">
<img class="custom-image" src="https://huggingface.co/EarthnDusk/Loras_2025/resolve/main/Arcane%20Illustrious%20Samples/image%20-%202025-04-08T191222.149.jpeg" alt="Preview">
</div>
</td>
<td>
<div class="custom-image-container">
<img class="custom-image" src="https://huggingface.co/EarthnDusk/Loras_2025/resolve/main/Resleeved%20Samples/image%20-%202025-04-08T141754.998.jpeg" alt="Preview">
</div>
</td>
</tr>
</table>
<details>
<summary> Supervised By </summary>
# Supervised by
**0FTH3N1GHT PRODUCTIONS**
More Information Coming Soon!
</details>
<details>
<summary>Support & Referrals</summary>
# Support
AI is our primary source of income. Your support is greatly appreciated!
| Platform | Link | Description |
|-----------------|----------------------------------------------------------------------|---------------------|
| **Ko-Fi** | [Duskfallcrew](https://ko-fi.com/duskfallcrew/) | Ko-Fi Duskfallcrew |
| **Ko-Fi** | [Earthnicity](https://ko-fi.com/earthnicity/) | Ko-Fi Earthnicity |
| **Ko-Fi** | [Rev. OTN Angel](https://ko-fi.com/OTNAngel/) | Ko-Fi Rev. OTN Angel |
| **Patreon** | [E&D Patreon](https://www.patreon.com/earthndusk) | E&D Patreon |
| **Merch** | [Merch Shop](https://duskfallcrew-shop.fourthwall.com/) | Merchandise |
| **Referral: Runpod** | [Runpod](https://runpod.io/?ref=yx1lcptf) | Runpod Referral |
| **Referral: VastAI**| [VastAI](https://cloud.vast.ai/?ref=70354) | VastAI Referral |
</details>
<details>
<summary>Connect with Earth & Dusk</summary>
# Social Media
| Platform | Link |
|-----------------|-------------------------------------------------------------------------|
| **Discord** | [E&D Discord](https://discord.gg/5t2kYxt7An) |
| **Discord (AI)**| [AI Discord](https://discord.gg/HhBSvM9gBY) |
| **Website** | [Website](https://end-media.org/) (Under Construction) |
| **Resources** | [Capsekai Resources](https://capsekai.carrd.co/) |
| **Subreddit** | [Subreddit](https://www.reddit.com/r/earthndusk/) |
| **YouTube** | [YouTube](https://www.youtube.com/channel/UCk7MGP7nrJz5awBSP75xmVw) |
| **TikTok** | [TikTok](https://www.tiktok.com/@duskfallcrew) |
| **Twitch** | [Twitch](https://twitch.tv/duskfallcrew) |
| **Instagram** | [Instagram](https://instagram.com/duskfallcrew) |
| **GitHub** | [Ktiseos-Nyx](https://github.com/Ktiseos-Nyx) |
</details>
<details>
<summary>Sponsors </summary>
# Partners & Sponsors
NOT ALL ARE PRESENTLY FINANCIALLY SPONSORING - These are also people who have sponsored us greatly in the past.
| Sponsor | Link |
|-------------------|--------------------------------------------|
| Pirate Diffusion | [Pirate Diffusion](https://www.piratediffusion.com/) |
| Yodayo/Moescape | [Yodayo/Moescape](https://moescape.ai/) |
Contact us for details on how to sponsor our content, or get our models on your platform!
</details>
<details>
<summary>Guidelines and Legal Information</summary>
# Legal & Guidelines
| Category | Guidelines |
|---------------|-------------------------------------------------------------------------------------------------------------------------------------------------------------|
| **Dos** | Use [XYPHER'S Tool](https://xypher7.github.io/lora-metadata-viewer/) to find metadata. Reuse, Recycle, and Merge! Credit creators & keep metadata. Convert to Diffusers, re-use, and re-integrate. |
| **Don'ts** | Re-upload our models *as is*. Use our content for illegal or immoral purposes. Claim our content as your own. Threaten or harm anyone. |
| **Legal** | Repositories fall under the **CREATIVE ML OPEN RAIL M FAMILY** license unless otherwise specified. Not for commercial redistribution. We are not legally responsible for outputs. |
| **Legal Names** | EARTH & DUSK MEDIA, Earth and Dusk Media, Ktiseos Nyx, Dusk/Duskfallcrew/The Duskfall Portal Crew/Dusky-crew, Earthnicity, The Introject Society. |
</details>
|
tashfinsami/model_bn
|
tashfinsami
| 2025-09-05T21:49:05Z | 6 | 0 |
diffusers
|
[
"diffusers",
"text-to-image",
"lora",
"diffusers-training",
"stable-diffusion",
"stable-diffusion-diffusers",
"base_model:Kardbord/stable-diffusion-v1-5-unsafe",
"base_model:adapter:Kardbord/stable-diffusion-v1-5-unsafe",
"license:creativeml-openrail-m",
"region:us"
] |
text-to-image
| 2025-08-28T07:20:36Z |
---
base_model: Kardbord/stable-diffusion-v1-5-unsafe
library_name: diffusers
license: creativeml-openrail-m
inference: true
instance_prompt: a derm photo of sks blue naevus lesion
tags:
- text-to-image
- diffusers
- lora
- diffusers-training
- stable-diffusion
- stable-diffusion-diffusers
- text-to-image
- diffusers
- lora
- diffusers-training
- stable-diffusion
- stable-diffusion-diffusers
---
<!-- This model card has been generated automatically according to the information the training script had access to. You
should probably proofread and complete it, then remove this comment. -->
# LoRA DreamBooth - tashfinsami/model_bn
These are LoRA adaption weights for Kardbord/stable-diffusion-v1-5-unsafe. The weights were trained on a derm photo of sks blue naevus lesion using [DreamBooth](https://dreambooth.github.io/). You can find some example images in the following.
LoRA for the text encoder was enabled: True.
## Intended uses & limitations
#### How to use
```python
# TODO: add an example code snippet for running this diffusion pipeline
```
#### Limitations and bias
[TODO: provide examples of latent issues and potential remediations]
## Training details
[TODO: describe the data used to train the model]
|
Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-Q5_K_R4-SPECIAL_SPLIT
|
Thireus
| 2025-09-05T21:48:33Z | 1 | 0 | null |
[
"gguf",
"arxiv:2505.23786",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-07-22T13:51:00Z |
---
license: mit
---
# DeepSeek-TNG-R1T2-Chimera
## 🤔 What is this [HuggingFace repository](https://huggingface.co/Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_SPLIT/) about?
This repository provides **GGUF-quantized tensors** for the DeepSeek-TNG-R1T2-Chimera model (official repo: https://huggingface.co/tngtech/DeepSeek-TNG-R1T2-Chimera). These GGUF shards are designed to be used with **Thireus’ GGUF Tool Suite** (https://gguf.thireus.com), a collection of tools that automatically finds the perplexity-optimal mix of quantizations for any given VRAM and RAM target. With the Tool Suite, you can generate and download custom quantization “recipes” effortlessly.
- 📖 Read more: https://github.com/Thireus/GGUF-Tool-Suite
- 🔍 Example quant mixes: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
- 🛠️ Create your own recipe: https://colab.research.google.com/github/Thireus/GGUF-Tool-Suite/blob/main/quant_recipe_pipeline.ipynb
- 📂 Browse available quant shards: https://huggingface.co/Thireus/collections
*tl;dr: Expand the details section below*
<details>
```
cd ~
# Make sure to install all ik_llama.cpp compilation dependencies...
apt install python3-dev python3-pip python3-venv python3-wheel python3-setuptools git acl netcat-openbsd cmake # pipx
# Obtain ik_llama's Thireus version - Windows builds available at https://github.com/Thireus/ik_llama.cpp/releases
git clone https://github.com/Thireus/ik_llama.cpp
cd ik_llama.cpp
git pull
# Build ik_llama.cpp
cmake -B build -DGGML_AVX=ON -DGGML_AVX2=ON -DLLAMA_CURL=OFF -DGGML_MAX_CONTEXTS=2048
cmake --build build --config Release -j16
cd ..
# Obtain Thireus' GGUF-Tool-Suite
git clone https://github.com/Thireus/GGUF-Tool-Suite
# Download model quant mix from recipe file:
cd GGUF-Tool-Suite
rm -f download.conf # Make sure to copy the relevant download.conf for the model before running quant_assign.py
cp -f models/DeepSeek-TNG-R1T2-Chimera/download.conf . # Use the download.conf of the chosen model
mkdir -p kitchen && cd kitchen
../quant_downloader.sh ../recipe_examples/ik_llama.cpp_recipes/DeepSeek-TNG-R1T2-Chimera.ROOT-3.0624bpw-3.3657ppl.238GB-GGUF_11GB-GPU_227GB-CPU.13549e6_1ac857a.recipe
# Launch ik_llama's llama-cli:
ulimit -n 9999 # Lifts "too many open files" limitation on Linux
~/ik_llama.cpp/build/bin/llama-cli \
-m DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_TENSOR-00001-of-01148.gguf \
-mla 3 -fa -amb 512 -fmoe -ctk f16 -c 4096 -ngl 99 \
-ot "blk\.(3|4|5|6)\.ffn_.*=CUDA0" \
-ot "blk\.(7|8|9|10)\.ffn_.*=CUDA1" \
-ot exps=CPU -b 2048 -ub 1024 --warmup-batch --no-mmap --threads 36 \
--main-gpu 0 \
-p '<｜begin▁of▁sentence｜><｜User｜>What is the solution of x+5=-2?<｜Assistant｜><think>\n'
```
</details>
---
## ❓ Why does this Tool Suite exist?
1. **Compatibility & Speed** – [unsloth](https://huggingface.co/unsloth)’s dynamic quants may not always work optimally with `ik_llama.cpp`.
2. **Custom Rig Fit** – No off-the-shelf GGUF model perfectly matched my VRAM/RAM setup, so I built a way to tailor models and leverage extra VRAM/RAM to reduce perplexity.
3. **Automated PPL-Optimal Quantization** – To my knowledge, there was no flexible, automated method to minimize perplexity for any bits-per-weight (bpw) target—so I created one with excellent results!
---
## 📊 How does it compare to other GGUFs?
Here’s how DeepSeek-R1-0528 quantized with **Thireus’ GGUF Tool Suite** stacks up against other quantizers (lower perplexity = better at equal or lower bpw):
> _Note: The `recipe_examples` files illustrate good recipes. The Tool Suite computes the optimal ppl/bpw curve for you — just specify your target RAM, VRAM, and quant types, and `quant_assign.py` finds the best mix._
More perplexity/bpw graphs for other supported models: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/ppl_graphs
*All PPL values are computed with the parameters `-ctk f16 -c 512 -b 4096 -ub 4096`. Changing any of these parameters will alter the PPL. In particular, reducing `-b 4096 -ub 4096` increases the PPL, while increasing them decreases the PPL.*
---
## 🚀 How do I get started?
Check out the [GGUF Tool Suite README](https://github.com/Thireus/GGUF-Tool-Suite) — focus on these sections:
1. ⚠️ **Requirements** – Which `ik_llama.cpp` (or `llama.cpp`) version to use and how to compile.
- Windows binaries (no patching needed) at: https://github.com/Thireus/ik_llama.cpp/releases
2. 📥 **Download Model Shards** – Use `quant_downloader.sh` to fetch GGUF shards from any recipe.
- Recipe examples: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
3. 🧠 **Run a Downloaded Model** – Sample usage with `llama-cli`.
4. 🛠️ **Generate a Custom Recipe** – Produce recipes tailored to your rig for optimal perplexity.
---
## ✅ Supported Models
Supported models are listed under `models/` in the [Tool Suite Github repo](https://github.com/Thireus/GGUF-Tool-Suite/tree/main/models). Presence of `ppl_results.csv` indicates official support and compatibility with `quant_assign.py`.
---
## 🤷♂️ Will I release pre-cooked GGUF files?
No, because I believe in **tailored quantization** for each user’s hardware. If you prefer ready-made shards, you are welcome to merge them via `llama-gguf-split --merge`, or request someone to publish them.
Instead, I prefer to share examples of recipes so users can see exactly how they were produced (command included inside these recipe files) and tweak them for their own rigs. The `quant_downloader.sh` script handles automatic fetching and verification of each shard. Recipes provided by [Ubergarm](https://huggingface.co/ubergarm) on his model cards are also compatible with `quant_downloader.sh`.
Users who don’t trust the GGUF shards on HuggingFace can also quantize their own by passing recipe lines to `llama-quantize --custom-q` ([see example](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/models/DeepSeek-R1-0528/DeepSeek-R1-0528-THIREUS-ANY-SPECIAL.sh#L482-L486)). Run `llama-quantize --help` to list compatible quants for `quant_assign.py`. This approach is especially useful if you prefer `llama.cpp` over `ik_llama.cpp`.
---
## 📦 What’s in this repository?
- **00001 GGUF header shard** – Contains metadata (tokens, chat template, tensor count, etc.). This metadata can be explored directly from the HuggingFace web interface after clicking on that shard.
- **Tensor shards** – Each shard holds one tensor; see `tensors.map` for names, quant types, sizes, SHA-256 hash, shard IDs, etc.
- **GPG-signed files** – `tensors.map` and header shard are signed with the key in [trusted-keys.asc](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/trusted-keys.asc) for tamper detection.
- **Security note** – Some papers about various ways to attack GGUFs and LLMs are available online, such as https://arxiv.org/abs/2505.23786, and there are also more classic security exploits like CVE-2024-23496 and CVE-2024-25664 through CVE-2024-25668. Only use GGUFs from reputable, trusted authors—or alternatively self-quantize—to avoid potential exploits.
---
## 💡 Pro Tips
You can download the BF16 model version to quantize your own shards:
```
mkdir kitchen
echo '.*=bf16' > kitchen/bf16.recipe
cd kitchen
../quant_downloader.sh bf16.recipe
```
Enjoy optimized quantization! 🎉
|
Stasonelison/blockassist-bc-howling_powerful_aardvark_1757108850
|
Stasonelison
| 2025-09-05T21:48:32Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"howling powerful aardvark",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:48:22Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- howling powerful aardvark
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-Q5_0-SPECIAL_SPLIT
|
Thireus
| 2025-09-05T21:44:56Z | 1 | 0 | null |
[
"gguf",
"arxiv:2505.23786",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-07-20T12:53:03Z |
---
license: mit
---
# DeepSeek-TNG-R1T2-Chimera
## 🤔 What is this [HuggingFace repository](https://huggingface.co/Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_SPLIT/) about?
This repository provides **GGUF-quantized tensors** for the DeepSeek-TNG-R1T2-Chimera model (official repo: https://huggingface.co/tngtech/DeepSeek-TNG-R1T2-Chimera). These GGUF shards are designed to be used with **Thireus’ GGUF Tool Suite** (https://gguf.thireus.com), a collection of tools that automatically finds the perplexity-optimal mix of quantizations for any given VRAM and RAM target. With the Tool Suite, you can generate and download custom quantization “recipes” effortlessly.
- 📖 Read more: https://github.com/Thireus/GGUF-Tool-Suite
- 🔍 Example quant mixes: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
- 🛠️ Create your own recipe: https://colab.research.google.com/github/Thireus/GGUF-Tool-Suite/blob/main/quant_recipe_pipeline.ipynb
- 📂 Browse available quant shards: https://huggingface.co/Thireus/collections
*tl;dr: Expand the details section below*
<details>
```
cd ~
# Make sure to install all ik_llama.cpp compilation dependencies...
apt install python3-dev python3-pip python3-venv python3-wheel python3-setuptools git acl netcat-openbsd cmake # pipx
# Obtain ik_llama's Thireus version - Windows builds available at https://github.com/Thireus/ik_llama.cpp/releases
git clone https://github.com/Thireus/ik_llama.cpp
cd ik_llama.cpp
git pull
# Build ik_llama.cpp
cmake -B build -DGGML_AVX=ON -DGGML_AVX2=ON -DLLAMA_CURL=OFF -DGGML_MAX_CONTEXTS=2048
cmake --build build --config Release -j16
cd ..
# Obtain Thireus' GGUF-Tool-Suite
git clone https://github.com/Thireus/GGUF-Tool-Suite
# Download model quant mix from recipe file:
cd GGUF-Tool-Suite
rm -f download.conf # Make sure to copy the relevant download.conf for the model before running quant_assign.py
cp -f models/DeepSeek-TNG-R1T2-Chimera/download.conf . # Use the download.conf of the chosen model
mkdir -p kitchen && cd kitchen
../quant_downloader.sh ../recipe_examples/ik_llama.cpp_recipes/DeepSeek-TNG-R1T2-Chimera.ROOT-3.0624bpw-3.3657ppl.238GB-GGUF_11GB-GPU_227GB-CPU.13549e6_1ac857a.recipe
# Launch ik_llama's llama-cli:
ulimit -n 9999 # Lifts "too many open files" limitation on Linux
~/ik_llama.cpp/build/bin/llama-cli \
-m DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_TENSOR-00001-of-01148.gguf \
-mla 3 -fa -amb 512 -fmoe -ctk f16 -c 4096 -ngl 99 \
-ot "blk\.(3|4|5|6)\.ffn_.*=CUDA0" \
-ot "blk\.(7|8|9|10)\.ffn_.*=CUDA1" \
-ot exps=CPU -b 2048 -ub 1024 --warmup-batch --no-mmap --threads 36 \
--main-gpu 0 \
-p '<｜begin▁of▁sentence｜><｜User｜>What is the solution of x+5=-2?<｜Assistant｜><think>\n'
```
</details>
---
## ❓ Why does this Tool Suite exist?
1. **Compatibility & Speed** – [unsloth](https://huggingface.co/unsloth)’s dynamic quants may not always work optimally with `ik_llama.cpp`.
2. **Custom Rig Fit** – No off-the-shelf GGUF model perfectly matched my VRAM/RAM setup, so I built a way to tailor models and leverage extra VRAM/RAM to reduce perplexity.
3. **Automated PPL-Optimal Quantization** – To my knowledge, there was no flexible, automated method to minimize perplexity for any bits-per-weight (bpw) target—so I created one with excellent results!
---
## 📊 How does it compare to other GGUFs?
Here’s how DeepSeek-R1-0528 quantized with **Thireus’ GGUF Tool Suite** stacks up against other quantizers (lower perplexity = better at equal or lower bpw):
> _Note: The `recipe_examples` files illustrate good recipes. The Tool Suite computes the optimal ppl/bpw curve for you — just specify your target RAM, VRAM, and quant types, and `quant_assign.py` finds the best mix._
More perplexity/bpw graphs for other supported models: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/ppl_graphs
*All PPL values are computed with the parameters `-ctk f16 -c 512 -b 4096 -ub 4096`. Changing any of these parameters will alter the PPL. In particular, reducing `-b 4096 -ub 4096` increases the PPL, while increasing them decreases the PPL.*
---
## 🚀 How do I get started?
Check out the [GGUF Tool Suite README](https://github.com/Thireus/GGUF-Tool-Suite) — focus on these sections:
1. ⚠️ **Requirements** – Which `ik_llama.cpp` (or `llama.cpp`) version to use and how to compile.
- Windows binaries (no patching needed) at: https://github.com/Thireus/ik_llama.cpp/releases
2. 📥 **Download Model Shards** – Use `quant_downloader.sh` to fetch GGUF shards from any recipe.
- Recipe examples: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
3. 🧠 **Run a Downloaded Model** – Sample usage with `llama-cli`.
4. 🛠️ **Generate a Custom Recipe** – Produce recipes tailored to your rig for optimal perplexity.
---
## ✅ Supported Models
Supported models are listed under `models/` in the [Tool Suite Github repo](https://github.com/Thireus/GGUF-Tool-Suite/tree/main/models). Presence of `ppl_results.csv` indicates official support and compatibility with `quant_assign.py`.
---
## 🤷♂️ Will I release pre-cooked GGUF files?
No, because I believe in **tailored quantization** for each user’s hardware. If you prefer ready-made shards, you are welcome to merge them via `llama-gguf-split --merge`, or request someone to publish them.
Instead, I prefer to share examples of recipes so users can see exactly how they were produced (command included inside these recipe files) and tweak them for their own rigs. The `quant_downloader.sh` script handles automatic fetching and verification of each shard. Recipes provided by [Ubergarm](https://huggingface.co/ubergarm) on his model cards are also compatible with `quant_downloader.sh`.
Users who don’t trust the GGUF shards on HuggingFace can also quantize their own by passing recipe lines to `llama-quantize --custom-q` ([see example](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/models/DeepSeek-R1-0528/DeepSeek-R1-0528-THIREUS-ANY-SPECIAL.sh#L482-L486)). Run `llama-quantize --help` to list compatible quants for `quant_assign.py`. This approach is especially useful if you prefer `llama.cpp` over `ik_llama.cpp`.
---
## 📦 What’s in this repository?
- **00001 GGUF header shard** – Contains metadata (tokens, chat template, tensor count, etc.). This metadata can be explored directly from the HuggingFace web interface after clicking on that shard.
- **Tensor shards** – Each shard holds one tensor; see `tensors.map` for names, quant types, sizes, SHA-256 hash, shard IDs, etc.
- **GPG-signed files** – `tensors.map` and header shard are signed with the key in [trusted-keys.asc](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/trusted-keys.asc) for tamper detection.
- **Security note** – Some papers about various ways to attack GGUFs and LLMs are available online, such as https://arxiv.org/abs/2505.23786, and there are also more classic security exploits like CVE-2024-23496 and CVE-2024-25664 through CVE-2024-25668. Only use GGUFs from reputable, trusted authors—or alternatively self-quantize—to avoid potential exploits.
---
## 💡 Pro Tips
You can download the BF16 model version to quantize your own shards:
```
mkdir kitchen
echo '.*=bf16' > kitchen/bf16.recipe
cd kitchen
../quant_downloader.sh bf16.recipe
```
Enjoy optimized quantization! 🎉
|
mradermacher/Seed-Coder-8B-Instruct-KTO-GGUF
|
mradermacher
| 2025-09-05T21:44:18Z | 0 | 0 |
transformers
|
[
"transformers",
"gguf",
"generated_from_trainer",
"trl",
"kto",
"en",
"base_model:willyli/Seed-Coder-8B-Instruct-KTO",
"base_model:quantized:willyli/Seed-Coder-8B-Instruct-KTO",
"endpoints_compatible",
"region:us",
"conversational"
] | null | 2025-09-05T20:23:07Z |
---
base_model: willyli/Seed-Coder-8B-Instruct-KTO
language:
- en
library_name: transformers
model_name: Seed-Coder-8B-Instruct-KTO
mradermacher:
readme_rev: 1
quantized_by: mradermacher
tags:
- generated_from_trainer
- trl
- kto
---
## About
<!-- ### quantize_version: 2 -->
<!-- ### output_tensor_quantised: 1 -->
<!-- ### convert_type: hf -->
<!-- ### vocab_type: -->
<!-- ### tags: -->
<!-- ### quants: x-f16 Q4_K_S Q2_K Q8_0 Q6_K Q3_K_M Q3_K_S Q3_K_L Q4_K_M Q5_K_S Q5_K_M IQ4_XS -->
<!-- ### quants_skip: -->
<!-- ### skip_mmproj: -->
static quants of https://huggingface.co/willyli/Seed-Coder-8B-Instruct-KTO
<!-- provided-files -->
***For a convenient overview and download list, visit our [model page for this model](https://hf.tst.eu/model#Seed-Coder-8B-Instruct-KTO-GGUF).***
weighted/imatrix quants seem not to be available (by me) at this time. If they do not show up a week or so after the static ones, I have probably not planned for them. Feel free to request them by opening a Community Discussion.
## Usage
If you are unsure how to use GGUF files, refer to one of [TheBloke's
READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
more details, including on how to concatenate multi-part files.
## Provided Quants
(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
| Link | Type | Size/GB | Notes |
|:-----|:-----|--------:|:------|
| [GGUF](https://huggingface.co/mradermacher/Seed-Coder-8B-Instruct-KTO-GGUF/resolve/main/Seed-Coder-8B-Instruct-KTO.Q2_K.gguf) | Q2_K | 3.4 | |
| [GGUF](https://huggingface.co/mradermacher/Seed-Coder-8B-Instruct-KTO-GGUF/resolve/main/Seed-Coder-8B-Instruct-KTO.Q3_K_S.gguf) | Q3_K_S | 3.9 | |
| [GGUF](https://huggingface.co/mradermacher/Seed-Coder-8B-Instruct-KTO-GGUF/resolve/main/Seed-Coder-8B-Instruct-KTO.Q3_K_M.gguf) | Q3_K_M | 4.3 | lower quality |
| [GGUF](https://huggingface.co/mradermacher/Seed-Coder-8B-Instruct-KTO-GGUF/resolve/main/Seed-Coder-8B-Instruct-KTO.Q3_K_L.gguf) | Q3_K_L | 4.6 | |
| [GGUF](https://huggingface.co/mradermacher/Seed-Coder-8B-Instruct-KTO-GGUF/resolve/main/Seed-Coder-8B-Instruct-KTO.IQ4_XS.gguf) | IQ4_XS | 4.7 | |
| [GGUF](https://huggingface.co/mradermacher/Seed-Coder-8B-Instruct-KTO-GGUF/resolve/main/Seed-Coder-8B-Instruct-KTO.Q4_K_S.gguf) | Q4_K_S | 4.9 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/Seed-Coder-8B-Instruct-KTO-GGUF/resolve/main/Seed-Coder-8B-Instruct-KTO.Q4_K_M.gguf) | Q4_K_M | 5.2 | fast, recommended |
| [GGUF](https://huggingface.co/mradermacher/Seed-Coder-8B-Instruct-KTO-GGUF/resolve/main/Seed-Coder-8B-Instruct-KTO.Q5_K_S.gguf) | Q5_K_S | 5.9 | |
| [GGUF](https://huggingface.co/mradermacher/Seed-Coder-8B-Instruct-KTO-GGUF/resolve/main/Seed-Coder-8B-Instruct-KTO.Q5_K_M.gguf) | Q5_K_M | 6.0 | |
| [GGUF](https://huggingface.co/mradermacher/Seed-Coder-8B-Instruct-KTO-GGUF/resolve/main/Seed-Coder-8B-Instruct-KTO.Q6_K.gguf) | Q6_K | 6.9 | very good quality |
| [GGUF](https://huggingface.co/mradermacher/Seed-Coder-8B-Instruct-KTO-GGUF/resolve/main/Seed-Coder-8B-Instruct-KTO.Q8_0.gguf) | Q8_0 | 8.9 | fast, best quality |
| [GGUF](https://huggingface.co/mradermacher/Seed-Coder-8B-Instruct-KTO-GGUF/resolve/main/Seed-Coder-8B-Instruct-KTO.f16.gguf) | f16 | 16.6 | 16 bpw, overkill |
Here is a handy graph by ikawrakow comparing some lower-quality quant
types (lower is better):
And here are Artefact2's thoughts on the matter:
https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
## FAQ / Model Request
See https://huggingface.co/mradermacher/model_requests for some answers to
questions you might have and/or if you want some other model quantized.
## Thanks
I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
me use its servers and providing upgrades to my workstation to enable
this work in my free time.
<!-- end -->
|
alaabh/Qwen3-8B-medical-merged-4bit
|
alaabh
| 2025-09-05T21:44:09Z | 0 | 0 |
transformers
|
[
"transformers",
"safetensors",
"qwen3",
"text-generation",
"text-generation-inference",
"unsloth",
"conversational",
"en",
"base_model:unsloth/Qwen3-8B-unsloth-bnb-4bit",
"base_model:quantized:unsloth/Qwen3-8B-unsloth-bnb-4bit",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"4-bit",
"bitsandbytes",
"region:us"
] |
text-generation
| 2025-09-05T21:42:33Z |
---
base_model: unsloth/Qwen3-8B-unsloth-bnb-4bit
tags:
- text-generation-inference
- transformers
- unsloth
- qwen3
license: apache-2.0
language:
- en
---
# Uploaded finetuned model
- **Developed by:** alaabh
- **License:** apache-2.0
- **Finetuned from model :** unsloth/Qwen3-8B-unsloth-bnb-4bit
This qwen3 model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|
Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-Q4_K_R4-SPECIAL_SPLIT
|
Thireus
| 2025-09-05T21:43:44Z | 1 | 0 | null |
[
"gguf",
"arxiv:2505.23786",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-07-22T10:34:14Z |
---
license: mit
---
# DeepSeek-TNG-R1T2-Chimera
## 🤔 What is this [HuggingFace repository](https://huggingface.co/Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_SPLIT/) about?
This repository provides **GGUF-quantized tensors** for the DeepSeek-TNG-R1T2-Chimera model (official repo: https://huggingface.co/tngtech/DeepSeek-TNG-R1T2-Chimera). These GGUF shards are designed to be used with **Thireus’ GGUF Tool Suite** (https://gguf.thireus.com), a collection of tools that automatically finds the perplexity-optimal mix of quantizations for any given VRAM and RAM target. With the Tool Suite, you can generate and download custom quantization “recipes” effortlessly.
- 📖 Read more: https://github.com/Thireus/GGUF-Tool-Suite
- 🔍 Example quant mixes: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
- 🛠️ Create your own recipe: https://colab.research.google.com/github/Thireus/GGUF-Tool-Suite/blob/main/quant_recipe_pipeline.ipynb
- 📂 Browse available quant shards: https://huggingface.co/Thireus/collections
*tl;dr: Expand the details section below*
<details>
```
cd ~
# Make sure to install all ik_llama.cpp compilation dependencies...
apt install python3-dev python3-pip python3-venv python3-wheel python3-setuptools git acl netcat-openbsd cmake # pipx
# Obtain ik_llama's Thireus version - Windows builds available at https://github.com/Thireus/ik_llama.cpp/releases
git clone https://github.com/Thireus/ik_llama.cpp
cd ik_llama.cpp
git pull
# Build ik_llama.cpp
cmake -B build -DGGML_AVX=ON -DGGML_AVX2=ON -DLLAMA_CURL=OFF -DGGML_MAX_CONTEXTS=2048
cmake --build build --config Release -j16
cd ..
# Obtain Thireus' GGUF-Tool-Suite
git clone https://github.com/Thireus/GGUF-Tool-Suite
# Download model quant mix from recipe file:
cd GGUF-Tool-Suite
rm -f download.conf # Make sure to copy the relevant download.conf for the model before running quant_assign.py
cp -f models/DeepSeek-TNG-R1T2-Chimera/download.conf . # Use the download.conf of the chosen model
mkdir -p kitchen && cd kitchen
../quant_downloader.sh ../recipe_examples/ik_llama.cpp_recipes/DeepSeek-TNG-R1T2-Chimera.ROOT-3.0624bpw-3.3657ppl.238GB-GGUF_11GB-GPU_227GB-CPU.13549e6_1ac857a.recipe
# Launch ik_llama's llama-cli:
ulimit -n 9999 # Lifts "too many open files" limitation on Linux
~/ik_llama.cpp/build/bin/llama-cli \
-m DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_TENSOR-00001-of-01148.gguf \
-mla 3 -fa -amb 512 -fmoe -ctk f16 -c 4096 -ngl 99 \
-ot "blk\.(3|4|5|6)\.ffn_.*=CUDA0" \
-ot "blk\.(7|8|9|10)\.ffn_.*=CUDA1" \
-ot exps=CPU -b 2048 -ub 1024 --warmup-batch --no-mmap --threads 36 \
--main-gpu 0 \
-p '<｜begin▁of▁sentence｜><｜User｜>What is the solution of x+5=-2?<｜Assistant｜><think>\n'
```
</details>
---
## ❓ Why does this Tool Suite exist?
1. **Compatibility & Speed** – [unsloth](https://huggingface.co/unsloth)’s dynamic quants may not always work optimally with `ik_llama.cpp`.
2. **Custom Rig Fit** – No off-the-shelf GGUF model perfectly matched my VRAM/RAM setup, so I built a way to tailor models and leverage extra VRAM/RAM to reduce perplexity.
3. **Automated PPL-Optimal Quantization** – To my knowledge, there was no flexible, automated method to minimize perplexity for any bits-per-weight (bpw) target—so I created one with excellent results!
---
## 📊 How does it compare to other GGUFs?
Here’s how DeepSeek-R1-0528 quantized with **Thireus’ GGUF Tool Suite** stacks up against other quantizers (lower perplexity = better at equal or lower bpw):
> _Note: The `recipe_examples` files illustrate good recipes. The Tool Suite computes the optimal ppl/bpw curve for you — just specify your target RAM, VRAM, and quant types, and `quant_assign.py` finds the best mix._
More perplexity/bpw graphs for other supported models: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/ppl_graphs
*All PPL values are computed with the parameters `-ctk f16 -c 512 -b 4096 -ub 4096`. Changing any of these parameters will alter the PPL. In particular, reducing `-b 4096 -ub 4096` increases the PPL, while increasing them decreases the PPL.*
---
## 🚀 How do I get started?
Check out the [GGUF Tool Suite README](https://github.com/Thireus/GGUF-Tool-Suite) — focus on these sections:
1. ⚠️ **Requirements** – Which `ik_llama.cpp` (or `llama.cpp`) version to use and how to compile.
- Windows binaries (no patching needed) at: https://github.com/Thireus/ik_llama.cpp/releases
2. 📥 **Download Model Shards** – Use `quant_downloader.sh` to fetch GGUF shards from any recipe.
- Recipe examples: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
3. 🧠 **Run a Downloaded Model** – Sample usage with `llama-cli`.
4. 🛠️ **Generate a Custom Recipe** – Produce recipes tailored to your rig for optimal perplexity.
---
## ✅ Supported Models
Supported models are listed under `models/` in the [Tool Suite Github repo](https://github.com/Thireus/GGUF-Tool-Suite/tree/main/models). Presence of `ppl_results.csv` indicates official support and compatibility with `quant_assign.py`.
---
## 🤷♂️ Will I release pre-cooked GGUF files?
No, because I believe in **tailored quantization** for each user’s hardware. If you prefer ready-made shards, you are welcome to merge them via `llama-gguf-split --merge`, or request someone to publish them.
Instead, I prefer to share examples of recipes so users can see exactly how they were produced (command included inside these recipe files) and tweak them for their own rigs. The `quant_downloader.sh` script handles automatic fetching and verification of each shard. Recipes provided by [Ubergarm](https://huggingface.co/ubergarm) on his model cards are also compatible with `quant_downloader.sh`.
Users who don’t trust the GGUF shards on HuggingFace can also quantize their own by passing recipe lines to `llama-quantize --custom-q` ([see example](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/models/DeepSeek-R1-0528/DeepSeek-R1-0528-THIREUS-ANY-SPECIAL.sh#L482-L486)). Run `llama-quantize --help` to list compatible quants for `quant_assign.py`. This approach is especially useful if you prefer `llama.cpp` over `ik_llama.cpp`.
---
## 📦 What’s in this repository?
- **00001 GGUF header shard** – Contains metadata (tokens, chat template, tensor count, etc.). This metadata can be explored directly from the HuggingFace web interface after clicking on that shard.
- **Tensor shards** – Each shard holds one tensor; see `tensors.map` for names, quant types, sizes, SHA-256 hash, shard IDs, etc.
- **GPG-signed files** – `tensors.map` and header shard are signed with the key in [trusted-keys.asc](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/trusted-keys.asc) for tamper detection.
- **Security note** – Some papers about various ways to attack GGUFs and LLMs are available online, such as https://arxiv.org/abs/2505.23786, and there are also more classic security exploits like CVE-2024-23496 and CVE-2024-25664 through CVE-2024-25668. Only use GGUFs from reputable, trusted authors—or alternatively self-quantize—to avoid potential exploits.
---
## 💡 Pro Tips
You can download the BF16 model version to quantize your own shards:
```
mkdir kitchen
echo '.*=bf16' > kitchen/bf16.recipe
cd kitchen
../quant_downloader.sh bf16.recipe
```
Enjoy optimized quantization! 🎉
|
Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-Q4_1-SPECIAL_SPLIT
|
Thireus
| 2025-09-05T21:42:30Z | 5 | 0 | null |
[
"gguf",
"arxiv:2505.23786",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-07-23T18:51:15Z |
---
license: mit
---
# DeepSeek-TNG-R1T2-Chimera
## 🤔 What is this [HuggingFace repository](https://huggingface.co/Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_SPLIT/) about?
This repository provides **GGUF-quantized tensors** for the DeepSeek-TNG-R1T2-Chimera model (official repo: https://huggingface.co/tngtech/DeepSeek-TNG-R1T2-Chimera). These GGUF shards are designed to be used with **Thireus’ GGUF Tool Suite** (https://gguf.thireus.com), a collection of tools that automatically finds the perplexity-optimal mix of quantizations for any given VRAM and RAM target. With the Tool Suite, you can generate and download custom quantization “recipes” effortlessly.
- 📖 Read more: https://github.com/Thireus/GGUF-Tool-Suite
- 🔍 Example quant mixes: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
- 🛠️ Create your own recipe: https://colab.research.google.com/github/Thireus/GGUF-Tool-Suite/blob/main/quant_recipe_pipeline.ipynb
- 📂 Browse available quant shards: https://huggingface.co/Thireus/collections
*tl;dr: Expand the details section below*
<details>
```
cd ~
# Make sure to install all ik_llama.cpp compilation dependencies...
apt install python3-dev python3-pip python3-venv python3-wheel python3-setuptools git acl netcat-openbsd cmake # pipx
# Obtain ik_llama's Thireus version - Windows builds available at https://github.com/Thireus/ik_llama.cpp/releases
git clone https://github.com/Thireus/ik_llama.cpp
cd ik_llama.cpp
git pull
# Build ik_llama.cpp
cmake -B build -DGGML_AVX=ON -DGGML_AVX2=ON -DLLAMA_CURL=OFF -DGGML_MAX_CONTEXTS=2048
cmake --build build --config Release -j16
cd ..
# Obtain Thireus' GGUF-Tool-Suite
git clone https://github.com/Thireus/GGUF-Tool-Suite
# Download model quant mix from recipe file:
cd GGUF-Tool-Suite
rm -f download.conf # Make sure to copy the relevant download.conf for the model before running quant_assign.py
cp -f models/DeepSeek-TNG-R1T2-Chimera/download.conf . # Use the download.conf of the chosen model
mkdir -p kitchen && cd kitchen
../quant_downloader.sh ../recipe_examples/ik_llama.cpp_recipes/DeepSeek-TNG-R1T2-Chimera.ROOT-3.0624bpw-3.3657ppl.238GB-GGUF_11GB-GPU_227GB-CPU.13549e6_1ac857a.recipe
# Launch ik_llama's llama-cli:
ulimit -n 9999 # Lifts "too many open files" limitation on Linux
~/ik_llama.cpp/build/bin/llama-cli \
-m DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_TENSOR-00001-of-01148.gguf \
-mla 3 -fa -amb 512 -fmoe -ctk f16 -c 4096 -ngl 99 \
-ot "blk\.(3|4|5|6)\.ffn_.*=CUDA0" \
-ot "blk\.(7|8|9|10)\.ffn_.*=CUDA1" \
-ot exps=CPU -b 2048 -ub 1024 --warmup-batch --no-mmap --threads 36 \
--main-gpu 0 \
-p '<｜begin▁of▁sentence｜><｜User｜>What is the solution of x+5=-2?<｜Assistant｜><think>\n'
```
</details>
---
## ❓ Why does this Tool Suite exist?
1. **Compatibility & Speed** – [unsloth](https://huggingface.co/unsloth)’s dynamic quants may not always work optimally with `ik_llama.cpp`.
2. **Custom Rig Fit** – No off-the-shelf GGUF model perfectly matched my VRAM/RAM setup, so I built a way to tailor models and leverage extra VRAM/RAM to reduce perplexity.
3. **Automated PPL-Optimal Quantization** – To my knowledge, there was no flexible, automated method to minimize perplexity for any bits-per-weight (bpw) target—so I created one with excellent results!
---
## 📊 How does it compare to other GGUFs?
Here’s how DeepSeek-R1-0528 quantized with **Thireus’ GGUF Tool Suite** stacks up against other quantizers (lower perplexity = better at equal or lower bpw):
> _Note: The `recipe_examples` files illustrate good recipes. The Tool Suite computes the optimal ppl/bpw curve for you — just specify your target RAM, VRAM, and quant types, and `quant_assign.py` finds the best mix._
More perplexity/bpw graphs for other supported models: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/ppl_graphs
*All PPL values are computed with the parameters `-ctk f16 -c 512 -b 4096 -ub 4096`. Changing any of these parameters will alter the PPL. In particular, reducing `-b 4096 -ub 4096` increases the PPL, while increasing them decreases the PPL.*
---
## 🚀 How do I get started?
Check out the [GGUF Tool Suite README](https://github.com/Thireus/GGUF-Tool-Suite) — focus on these sections:
1. ⚠️ **Requirements** – Which `ik_llama.cpp` (or `llama.cpp`) version to use and how to compile.
- Windows binaries (no patching needed) at: https://github.com/Thireus/ik_llama.cpp/releases
2. 📥 **Download Model Shards** – Use `quant_downloader.sh` to fetch GGUF shards from any recipe.
- Recipe examples: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
3. 🧠 **Run a Downloaded Model** – Sample usage with `llama-cli`.
4. 🛠️ **Generate a Custom Recipe** – Produce recipes tailored to your rig for optimal perplexity.
---
## ✅ Supported Models
Supported models are listed under `models/` in the [Tool Suite Github repo](https://github.com/Thireus/GGUF-Tool-Suite/tree/main/models). Presence of `ppl_results.csv` indicates official support and compatibility with `quant_assign.py`.
---
## 🤷♂️ Will I release pre-cooked GGUF files?
No, because I believe in **tailored quantization** for each user’s hardware. If you prefer ready-made shards, you are welcome to merge them via `llama-gguf-split --merge`, or request someone to publish them.
Instead, I prefer to share examples of recipes so users can see exactly how they were produced (command included inside these recipe files) and tweak them for their own rigs. The `quant_downloader.sh` script handles automatic fetching and verification of each shard. Recipes provided by [Ubergarm](https://huggingface.co/ubergarm) on his model cards are also compatible with `quant_downloader.sh`.
Users who don’t trust the GGUF shards on HuggingFace can also quantize their own by passing recipe lines to `llama-quantize --custom-q` ([see example](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/models/DeepSeek-R1-0528/DeepSeek-R1-0528-THIREUS-ANY-SPECIAL.sh#L482-L486)). Run `llama-quantize --help` to list compatible quants for `quant_assign.py`. This approach is especially useful if you prefer `llama.cpp` over `ik_llama.cpp`.
---
## 📦 What’s in this repository?
- **00001 GGUF header shard** – Contains metadata (tokens, chat template, tensor count, etc.). This metadata can be explored directly from the HuggingFace web interface after clicking on that shard.
- **Tensor shards** – Each shard holds one tensor; see `tensors.map` for names, quant types, sizes, SHA-256 hash, shard IDs, etc.
- **GPG-signed files** – `tensors.map` and header shard are signed with the key in [trusted-keys.asc](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/trusted-keys.asc) for tamper detection.
- **Security note** – Some papers about various ways to attack GGUFs and LLMs are available online, such as https://arxiv.org/abs/2505.23786, and there are also more classic security exploits like CVE-2024-23496 and CVE-2024-25664 through CVE-2024-25668. Only use GGUFs from reputable, trusted authors—or alternatively self-quantize—to avoid potential exploits.
---
## 💡 Pro Tips
You can download the BF16 model version to quantize your own shards:
```
mkdir kitchen
echo '.*=bf16' > kitchen/bf16.recipe
cd kitchen
../quant_downloader.sh bf16.recipe
```
Enjoy optimized quantization! 🎉
|
alaabh/Qwen3-8B-medical-merged-16bit
|
alaabh
| 2025-09-05T21:42:29Z | 0 | 0 |
transformers
|
[
"transformers",
"safetensors",
"qwen3",
"text-generation",
"text-generation-inference",
"unsloth",
"conversational",
"en",
"base_model:unsloth/Qwen3-8B-unsloth-bnb-4bit",
"base_model:finetune:unsloth/Qwen3-8B-unsloth-bnb-4bit",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] |
text-generation
| 2025-09-05T20:58:32Z |
---
base_model: unsloth/Qwen3-8B-unsloth-bnb-4bit
tags:
- text-generation-inference
- transformers
- unsloth
- qwen3
license: apache-2.0
language:
- en
---
# Uploaded finetuned model
- **Developed by:** alaabh
- **License:** apache-2.0
- **Finetuned from model :** unsloth/Qwen3-8B-unsloth-bnb-4bit
This qwen3 model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|
Stasonelison/blockassist-bc-howling_powerful_aardvark_1757108483
|
Stasonelison
| 2025-09-05T21:42:23Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"howling powerful aardvark",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:42:13Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- howling powerful aardvark
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
yopings/blockassist-bc-barky_rangy_tapir_1757108476
|
yopings
| 2025-09-05T21:42:06Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"barky rangy tapir",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:41:48Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- barky rangy tapir
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-Q4_0_R8-SPECIAL_SPLIT
|
Thireus
| 2025-09-05T21:41:16Z | 12 | 0 | null |
[
"gguf",
"arxiv:2505.23786",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-07-22T10:34:01Z |
---
license: mit
---
# DeepSeek-TNG-R1T2-Chimera
## 🤔 What is this [HuggingFace repository](https://huggingface.co/Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_SPLIT/) about?
This repository provides **GGUF-quantized tensors** for the DeepSeek-TNG-R1T2-Chimera model (official repo: https://huggingface.co/tngtech/DeepSeek-TNG-R1T2-Chimera). These GGUF shards are designed to be used with **Thireus’ GGUF Tool Suite** (https://gguf.thireus.com), a collection of tools that automatically finds the perplexity-optimal mix of quantizations for any given VRAM and RAM target. With the Tool Suite, you can generate and download custom quantization “recipes” effortlessly.
- 📖 Read more: https://github.com/Thireus/GGUF-Tool-Suite
- 🔍 Example quant mixes: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
- 🛠️ Create your own recipe: https://colab.research.google.com/github/Thireus/GGUF-Tool-Suite/blob/main/quant_recipe_pipeline.ipynb
- 📂 Browse available quant shards: https://huggingface.co/Thireus/collections
*tl;dr: Expand the details section below*
<details>
```
cd ~
# Make sure to install all ik_llama.cpp compilation dependencies...
apt install python3-dev python3-pip python3-venv python3-wheel python3-setuptools git acl netcat-openbsd cmake # pipx
# Obtain ik_llama's Thireus version - Windows builds available at https://github.com/Thireus/ik_llama.cpp/releases
git clone https://github.com/Thireus/ik_llama.cpp
cd ik_llama.cpp
git pull
# Build ik_llama.cpp
cmake -B build -DGGML_AVX=ON -DGGML_AVX2=ON -DLLAMA_CURL=OFF -DGGML_MAX_CONTEXTS=2048
cmake --build build --config Release -j16
cd ..
# Obtain Thireus' GGUF-Tool-Suite
git clone https://github.com/Thireus/GGUF-Tool-Suite
# Download model quant mix from recipe file:
cd GGUF-Tool-Suite
rm -f download.conf # Make sure to copy the relevant download.conf for the model before running quant_assign.py
cp -f models/DeepSeek-TNG-R1T2-Chimera/download.conf . # Use the download.conf of the chosen model
mkdir -p kitchen && cd kitchen
../quant_downloader.sh ../recipe_examples/ik_llama.cpp_recipes/DeepSeek-TNG-R1T2-Chimera.ROOT-3.0624bpw-3.3657ppl.238GB-GGUF_11GB-GPU_227GB-CPU.13549e6_1ac857a.recipe
# Launch ik_llama's llama-cli:
ulimit -n 9999 # Lifts "too many open files" limitation on Linux
~/ik_llama.cpp/build/bin/llama-cli \
-m DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_TENSOR-00001-of-01148.gguf \
-mla 3 -fa -amb 512 -fmoe -ctk f16 -c 4096 -ngl 99 \
-ot "blk\.(3|4|5|6)\.ffn_.*=CUDA0" \
-ot "blk\.(7|8|9|10)\.ffn_.*=CUDA1" \
-ot exps=CPU -b 2048 -ub 1024 --warmup-batch --no-mmap --threads 36 \
--main-gpu 0 \
-p '<｜begin▁of▁sentence｜><｜User｜>What is the solution of x+5=-2?<｜Assistant｜><think>\n'
```
</details>
---
## ❓ Why does this Tool Suite exist?
1. **Compatibility & Speed** – [unsloth](https://huggingface.co/unsloth)’s dynamic quants may not always work optimally with `ik_llama.cpp`.
2. **Custom Rig Fit** – No off-the-shelf GGUF model perfectly matched my VRAM/RAM setup, so I built a way to tailor models and leverage extra VRAM/RAM to reduce perplexity.
3. **Automated PPL-Optimal Quantization** – To my knowledge, there was no flexible, automated method to minimize perplexity for any bits-per-weight (bpw) target—so I created one with excellent results!
---
## 📊 How does it compare to other GGUFs?
Here’s how DeepSeek-R1-0528 quantized with **Thireus’ GGUF Tool Suite** stacks up against other quantizers (lower perplexity = better at equal or lower bpw):
> _Note: The `recipe_examples` files illustrate good recipes. The Tool Suite computes the optimal ppl/bpw curve for you — just specify your target RAM, VRAM, and quant types, and `quant_assign.py` finds the best mix._
More perplexity/bpw graphs for other supported models: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/ppl_graphs
*All PPL values are computed with the parameters `-ctk f16 -c 512 -b 4096 -ub 4096`. Changing any of these parameters will alter the PPL. In particular, reducing `-b 4096 -ub 4096` increases the PPL, while increasing them decreases the PPL.*
---
## 🚀 How do I get started?
Check out the [GGUF Tool Suite README](https://github.com/Thireus/GGUF-Tool-Suite) — focus on these sections:
1. ⚠️ **Requirements** – Which `ik_llama.cpp` (or `llama.cpp`) version to use and how to compile.
- Windows binaries (no patching needed) at: https://github.com/Thireus/ik_llama.cpp/releases
2. 📥 **Download Model Shards** – Use `quant_downloader.sh` to fetch GGUF shards from any recipe.
- Recipe examples: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
3. 🧠 **Run a Downloaded Model** – Sample usage with `llama-cli`.
4. 🛠️ **Generate a Custom Recipe** – Produce recipes tailored to your rig for optimal perplexity.
---
## ✅ Supported Models
Supported models are listed under `models/` in the [Tool Suite Github repo](https://github.com/Thireus/GGUF-Tool-Suite/tree/main/models). Presence of `ppl_results.csv` indicates official support and compatibility with `quant_assign.py`.
---
## 🤷♂️ Will I release pre-cooked GGUF files?
No, because I believe in **tailored quantization** for each user’s hardware. If you prefer ready-made shards, you are welcome to merge them via `llama-gguf-split --merge`, or request someone to publish them.
Instead, I prefer to share examples of recipes so users can see exactly how they were produced (command included inside these recipe files) and tweak them for their own rigs. The `quant_downloader.sh` script handles automatic fetching and verification of each shard. Recipes provided by [Ubergarm](https://huggingface.co/ubergarm) on his model cards are also compatible with `quant_downloader.sh`.
Users who don’t trust the GGUF shards on HuggingFace can also quantize their own by passing recipe lines to `llama-quantize --custom-q` ([see example](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/models/DeepSeek-R1-0528/DeepSeek-R1-0528-THIREUS-ANY-SPECIAL.sh#L482-L486)). Run `llama-quantize --help` to list compatible quants for `quant_assign.py`. This approach is especially useful if you prefer `llama.cpp` over `ik_llama.cpp`.
---
## 📦 What’s in this repository?
- **00001 GGUF header shard** – Contains metadata (tokens, chat template, tensor count, etc.). This metadata can be explored directly from the HuggingFace web interface after clicking on that shard.
- **Tensor shards** – Each shard holds one tensor; see `tensors.map` for names, quant types, sizes, SHA-256 hash, shard IDs, etc.
- **GPG-signed files** – `tensors.map` and header shard are signed with the key in [trusted-keys.asc](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/trusted-keys.asc) for tamper detection.
- **Security note** – Some papers about various ways to attack GGUFs and LLMs are available online, such as https://arxiv.org/abs/2505.23786, and there are also more classic security exploits like CVE-2024-23496 and CVE-2024-25664 through CVE-2024-25668. Only use GGUFs from reputable, trusted authors—or alternatively self-quantize—to avoid potential exploits.
---
## 💡 Pro Tips
You can download the BF16 model version to quantize your own shards:
```
mkdir kitchen
echo '.*=bf16' > kitchen/bf16.recipe
cd kitchen
../quant_downloader.sh bf16.recipe
```
Enjoy optimized quantization! 🎉
|
bah63843/blockassist-bc-plump_fast_antelope_1757108339
|
bah63843
| 2025-09-05T21:39:49Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"plump fast antelope",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:39:41Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- plump fast antelope
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
qgallouedec/Qwen3-8B-SFT-20250905191104
|
qgallouedec
| 2025-09-05T21:39:06Z | 0 | 0 |
transformers
|
[
"transformers",
"safetensors",
"generated_from_trainer",
"sft",
"hf_jobs",
"trl",
"base_model:Qwen/Qwen3-8B",
"base_model:finetune:Qwen/Qwen3-8B",
"endpoints_compatible",
"region:us"
] | null | 2025-09-05T19:11:59Z |
---
base_model: Qwen/Qwen3-8B
library_name: transformers
model_name: Qwen3-8B-SFT-20250905191104
tags:
- generated_from_trainer
- sft
- hf_jobs
- trl
licence: license
---
# Model Card for Qwen3-8B-SFT-20250905191104
This model is a fine-tuned version of [Qwen/Qwen3-8B](https://huggingface.co/Qwen/Qwen3-8B).
It has been trained using [TRL](https://github.com/huggingface/trl).
## Quick start
```python
from transformers import pipeline
question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
generator = pipeline("text-generation", model="qgallouedec/Qwen3-8B-SFT-20250905191104", device="cuda")
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
print(output["generated_text"])
```
## Training procedure
This model was trained with SFT.
### Framework versions
- TRL: 0.23.0.dev0
- Transformers: 4.56.1
- Pytorch: 2.8.0+cu128
- Datasets: 4.0.0
- Tokenizers: 0.22.0
## Citations
Cite TRL as:
```bibtex
@misc{vonwerra2022trl,
title = {{TRL: Transformer Reinforcement Learning}},
author = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallou{\'e}dec},
year = 2020,
journal = {GitHub repository},
publisher = {GitHub},
howpublished = {\url{https://github.com/huggingface/trl}}
}
```
|
Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-Q3_K_R4-SPECIAL_SPLIT
|
Thireus
| 2025-09-05T21:38:51Z | 1 | 0 | null |
[
"gguf",
"arxiv:2505.23786",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-07-22T10:33:33Z |
---
license: mit
---
# DeepSeek-TNG-R1T2-Chimera
## 🤔 What is this [HuggingFace repository](https://huggingface.co/Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_SPLIT/) about?
This repository provides **GGUF-quantized tensors** for the DeepSeek-TNG-R1T2-Chimera model (official repo: https://huggingface.co/tngtech/DeepSeek-TNG-R1T2-Chimera). These GGUF shards are designed to be used with **Thireus’ GGUF Tool Suite** (https://gguf.thireus.com), a collection of tools that automatically finds the perplexity-optimal mix of quantizations for any given VRAM and RAM target. With the Tool Suite, you can generate and download custom quantization “recipes” effortlessly.
- 📖 Read more: https://github.com/Thireus/GGUF-Tool-Suite
- 🔍 Example quant mixes: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
- 🛠️ Create your own recipe: https://colab.research.google.com/github/Thireus/GGUF-Tool-Suite/blob/main/quant_recipe_pipeline.ipynb
- 📂 Browse available quant shards: https://huggingface.co/Thireus/collections
*tl;dr: Expand the details section below*
<details>
```
cd ~
# Make sure to install all ik_llama.cpp compilation dependencies...
apt install python3-dev python3-pip python3-venv python3-wheel python3-setuptools git acl netcat-openbsd cmake # pipx
# Obtain ik_llama's Thireus version - Windows builds available at https://github.com/Thireus/ik_llama.cpp/releases
git clone https://github.com/Thireus/ik_llama.cpp
cd ik_llama.cpp
git pull
# Build ik_llama.cpp
cmake -B build -DGGML_AVX=ON -DGGML_AVX2=ON -DLLAMA_CURL=OFF -DGGML_MAX_CONTEXTS=2048
cmake --build build --config Release -j16
cd ..
# Obtain Thireus' GGUF-Tool-Suite
git clone https://github.com/Thireus/GGUF-Tool-Suite
# Download model quant mix from recipe file:
cd GGUF-Tool-Suite
rm -f download.conf # Make sure to copy the relevant download.conf for the model before running quant_assign.py
cp -f models/DeepSeek-TNG-R1T2-Chimera/download.conf . # Use the download.conf of the chosen model
mkdir -p kitchen && cd kitchen
../quant_downloader.sh ../recipe_examples/ik_llama.cpp_recipes/DeepSeek-TNG-R1T2-Chimera.ROOT-3.0624bpw-3.3657ppl.238GB-GGUF_11GB-GPU_227GB-CPU.13549e6_1ac857a.recipe
# Launch ik_llama's llama-cli:
ulimit -n 9999 # Lifts "too many open files" limitation on Linux
~/ik_llama.cpp/build/bin/llama-cli \
-m DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_TENSOR-00001-of-01148.gguf \
-mla 3 -fa -amb 512 -fmoe -ctk f16 -c 4096 -ngl 99 \
-ot "blk\.(3|4|5|6)\.ffn_.*=CUDA0" \
-ot "blk\.(7|8|9|10)\.ffn_.*=CUDA1" \
-ot exps=CPU -b 2048 -ub 1024 --warmup-batch --no-mmap --threads 36 \
--main-gpu 0 \
-p '<｜begin▁of▁sentence｜><｜User｜>What is the solution of x+5=-2?<｜Assistant｜><think>\n'
```
</details>
---
## ❓ Why does this Tool Suite exist?
1. **Compatibility & Speed** – [unsloth](https://huggingface.co/unsloth)’s dynamic quants may not always work optimally with `ik_llama.cpp`.
2. **Custom Rig Fit** – No off-the-shelf GGUF model perfectly matched my VRAM/RAM setup, so I built a way to tailor models and leverage extra VRAM/RAM to reduce perplexity.
3. **Automated PPL-Optimal Quantization** – To my knowledge, there was no flexible, automated method to minimize perplexity for any bits-per-weight (bpw) target—so I created one with excellent results!
---
## 📊 How does it compare to other GGUFs?
Here’s how DeepSeek-R1-0528 quantized with **Thireus’ GGUF Tool Suite** stacks up against other quantizers (lower perplexity = better at equal or lower bpw):
> _Note: The `recipe_examples` files illustrate good recipes. The Tool Suite computes the optimal ppl/bpw curve for you — just specify your target RAM, VRAM, and quant types, and `quant_assign.py` finds the best mix._
More perplexity/bpw graphs for other supported models: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/ppl_graphs
*All PPL values are computed with the parameters `-ctk f16 -c 512 -b 4096 -ub 4096`. Changing any of these parameters will alter the PPL. In particular, reducing `-b 4096 -ub 4096` increases the PPL, while increasing them decreases the PPL.*
---
## 🚀 How do I get started?
Check out the [GGUF Tool Suite README](https://github.com/Thireus/GGUF-Tool-Suite) — focus on these sections:
1. ⚠️ **Requirements** – Which `ik_llama.cpp` (or `llama.cpp`) version to use and how to compile.
- Windows binaries (no patching needed) at: https://github.com/Thireus/ik_llama.cpp/releases
2. 📥 **Download Model Shards** – Use `quant_downloader.sh` to fetch GGUF shards from any recipe.
- Recipe examples: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
3. 🧠 **Run a Downloaded Model** – Sample usage with `llama-cli`.
4. 🛠️ **Generate a Custom Recipe** – Produce recipes tailored to your rig for optimal perplexity.
---
## ✅ Supported Models
Supported models are listed under `models/` in the [Tool Suite Github repo](https://github.com/Thireus/GGUF-Tool-Suite/tree/main/models). Presence of `ppl_results.csv` indicates official support and compatibility with `quant_assign.py`.
---
## 🤷♂️ Will I release pre-cooked GGUF files?
No, because I believe in **tailored quantization** for each user’s hardware. If you prefer ready-made shards, you are welcome to merge them via `llama-gguf-split --merge`, or request someone to publish them.
Instead, I prefer to share examples of recipes so users can see exactly how they were produced (command included inside these recipe files) and tweak them for their own rigs. The `quant_downloader.sh` script handles automatic fetching and verification of each shard. Recipes provided by [Ubergarm](https://huggingface.co/ubergarm) on his model cards are also compatible with `quant_downloader.sh`.
Users who don’t trust the GGUF shards on HuggingFace can also quantize their own by passing recipe lines to `llama-quantize --custom-q` ([see example](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/models/DeepSeek-R1-0528/DeepSeek-R1-0528-THIREUS-ANY-SPECIAL.sh#L482-L486)). Run `llama-quantize --help` to list compatible quants for `quant_assign.py`. This approach is especially useful if you prefer `llama.cpp` over `ik_llama.cpp`.
---
## 📦 What’s in this repository?
- **00001 GGUF header shard** – Contains metadata (tokens, chat template, tensor count, etc.). This metadata can be explored directly from the HuggingFace web interface after clicking on that shard.
- **Tensor shards** – Each shard holds one tensor; see `tensors.map` for names, quant types, sizes, SHA-256 hash, shard IDs, etc.
- **GPG-signed files** – `tensors.map` and header shard are signed with the key in [trusted-keys.asc](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/trusted-keys.asc) for tamper detection.
- **Security note** – Some papers about various ways to attack GGUFs and LLMs are available online, such as https://arxiv.org/abs/2505.23786, and there are also more classic security exploits like CVE-2024-23496 and CVE-2024-25664 through CVE-2024-25668. Only use GGUFs from reputable, trusted authors—or alternatively self-quantize—to avoid potential exploits.
---
## 💡 Pro Tips
You can download the BF16 model version to quantize your own shards:
```
mkdir kitchen
echo '.*=bf16' > kitchen/bf16.recipe
cd kitchen
../quant_downloader.sh bf16.recipe
```
Enjoy optimized quantization! 🎉
|
Stasonelison/blockassist-bc-howling_powerful_aardvark_1757108230
|
Stasonelison
| 2025-09-05T21:38:02Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"howling powerful aardvark",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:37:51Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- howling powerful aardvark
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
cwayneconnor/blockassist-bc-mute_loud_lynx_1757107943
|
cwayneconnor
| 2025-09-05T21:36:38Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"mute loud lynx",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:33:49Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- mute loud lynx
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
psheth2s/wav2vec2-tess-emotion
|
psheth2s
| 2025-09-05T21:36:13Z | 0 | 0 |
transformers
|
[
"transformers",
"safetensors",
"wav2vec2",
"audio-classification",
"arxiv:1910.09700",
"endpoints_compatible",
"region:us"
] |
audio-classification
| 2025-09-05T21:35:49Z |
---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed]
|
rocktanmay2012/blockassist-bc-bold_placid_barracuda_1757108048
|
rocktanmay2012
| 2025-09-05T21:34:46Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"bold placid barracuda",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:34:35Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- bold placid barracuda
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
Rakancorle1/qwen2.5-32B_Instruct_0905_policy_traj_30k_full
|
Rakancorle1
| 2025-09-05T21:34:19Z | 0 | 0 |
transformers
|
[
"transformers",
"safetensors",
"qwen2",
"text-generation",
"llama-factory",
"full",
"generated_from_trainer",
"conversational",
"base_model:Qwen/Qwen2.5-32B-Instruct",
"base_model:finetune:Qwen/Qwen2.5-32B-Instruct",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] |
text-generation
| 2025-09-05T19:50:08Z |
---
library_name: transformers
license: apache-2.0
base_model: Qwen/Qwen2.5-32B-Instruct
tags:
- llama-factory
- full
- generated_from_trainer
model-index:
- name: qwen2.5-32B_Instruct_0905_policy_traj_30k_full
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# qwen2.5-32B_Instruct_0905_policy_traj_30k_full
This model is a fine-tuned version of [Qwen/Qwen2.5-32B-Instruct](https://huggingface.co/Qwen/Qwen2.5-32B-Instruct) on the Policy_Traj_0826_30k_train dataset.
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-06
- train_batch_size: 2
- eval_batch_size: 8
- seed: 42
- distributed_type: multi-GPU
- num_devices: 8
- gradient_accumulation_steps: 16
- total_train_batch_size: 256
- total_eval_batch_size: 64
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.03
- num_epochs: 3.0
### Training results
### Framework versions
- Transformers 4.55.0
- Pytorch 2.7.0+cu126
- Datasets 3.6.0
- Tokenizers 0.21.1
|
bah63843/blockassist-bc-plump_fast_antelope_1757107990
|
bah63843
| 2025-09-05T21:34:13Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"plump fast antelope",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:33:52Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- plump fast antelope
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
Regan0323/Llama-3.2-3B-Instruct-full
|
Regan0323
| 2025-09-05T21:33:55Z | 0 | 0 |
transformers
|
[
"transformers",
"tensorboard",
"safetensors",
"generated_from_trainer",
"sft",
"trl",
"base_model:meta-llama/Llama-3.2-3B-Instruct",
"base_model:finetune:meta-llama/Llama-3.2-3B-Instruct",
"endpoints_compatible",
"region:us"
] | null | 2025-09-05T21:32:42Z |
---
base_model: meta-llama/Llama-3.2-3B-Instruct
library_name: transformers
model_name: Llama-3.2-3B-Instruct-full
tags:
- generated_from_trainer
- sft
- trl
licence: license
---
# Model Card for Llama-3.2-3B-Instruct-full
This model is a fine-tuned version of [meta-llama/Llama-3.2-3B-Instruct](https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct).
It has been trained using [TRL](https://github.com/huggingface/trl).
## Quick start
```python
from transformers import pipeline
question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
generator = pipeline("text-generation", model="Regan0323/Llama-3.2-3B-Instruct-full", device="cuda")
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
print(output["generated_text"])
```
## Training procedure
This model was trained with SFT.
### Framework versions
- TRL: 0.22.2
- Transformers: 4.56.1
- Pytorch: 2.8.0+cu126
- Datasets: 4.0.0
- Tokenizers: 0.22.0
## Citations
Cite TRL as:
```bibtex
@misc{vonwerra2022trl,
title = {{TRL: Transformer Reinforcement Learning}},
author = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallou{\'e}dec},
year = 2020,
journal = {GitHub repository},
publisher = {GitHub},
howpublished = {\url{https://github.com/huggingface/trl}}
}
```
|
klmdr22/blockassist-bc-wild_loud_newt_1757107974
|
klmdr22
| 2025-09-05T21:33:35Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"wild loud newt",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:33:32Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- wild loud newt
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
dilip025/llama-2-7b
|
dilip025
| 2025-09-05T21:33:25Z | 15 | 0 |
transformers
|
[
"transformers",
"pytorch",
"llama",
"text-generation",
"facebook",
"meta",
"llama-2",
"en",
"base_model:meta-llama/Llama-2-7b-chat-hf",
"base_model:finetune:meta-llama/Llama-2-7b-chat-hf",
"license:llama2",
"autotrain_compatible",
"text-generation-inference",
"region:us"
] |
text-generation
| 2024-03-02T17:03:29Z |
---
language:
- en
license: llama2
tags:
- facebook
- meta
- pytorch
- llama
- llama-2
model_name: Llama 2 7B Chat
arxiv: 2307.09288
base_model: meta-llama/Llama-2-7b-chat-hf
inference: false
model_creator: Meta Llama 2
model_type: llama
pipeline_tag: text-generation
prompt_template: '[INST] <<SYS>>
You are NutriLife chatbot, you are going to get questions related to food, nutrition, health, and diet by the users from Nepal. Answer them very shortly and accurately if the message is only about food, nutrition, and diet. Otherwise, ignore.
<</SYS>>
{prompt}[/INST]
'
quantized_by: Dilip Pokhrel
---
<hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
<!-- header end -->
# Llama 2 7B Chat -- Food and Nutrition
<br>
- Model creator: [Meta Llama 2]
<br>
- Original model: [Llama 2 7B Chat] <a href="https://huggingface.co/meta-llama/Llama-2-7b-chat-hf">Original Model</a>
<br>
- Fine Tuned by: [Dilip Pokhrel] <a href="https://dilippokhrel.com.np">Profile</a>
#### Simple example code to load one of these GGUF models
```python
# Load model directly or use qunatization technique if you have low gpu ram
from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("dilip025/llama-2-7b")
model = AutoModelForCausalLM.from_pretrained("dilip025/llama-2-7b")
system_message = 'You are NutriLife chatbot, you are going to get questions related to food, nutrition, health, and diet by the users from Nepal. Answer them very shortly and accurately if the message is only about food, nutrition, and diet. Otherwise, ignore.'
prompt = f"[INST] <<SYS>>\n{system_message}\n<</SYS>>\n\n Tell me some of the famous Nepali food recipes [/INST]"
num_new_tokens = 200 # Change to the number of new tokens you want to generate
# Count the number of tokens in the prompt
num_prompt_tokens = len(tokenizer(prompt)['input_ids'])
# Calculate the maximum length for the generation
max_length = num_prompt_tokens + num_new_tokens
gen = pipeline('text-generation', model=model, tokenizer=tokenizer, max_length=max_length)
result = gen(prompt)
print(result[0]['generated_text'].replace(prompt, ''))
```
## Ethical Considerations and Limitations
Llama 2 is a new technology that carries risks with use. Testing conducted to date has been in English, and has not covered, nor could it cover all scenarios. For these reasons, as with all LLMs, Llama 2’s potential outputs cannot be predicted in advance, and the model may in some instances produce inaccurate, biased or other objectionable responses to user prompts. Therefore, before deploying any applications of Llama 2, developers should perform safety testing and tuning tailored to their specific applications of the model.
Please see the Responsible Use Guide available at [https://ai.meta.com/llama/responsible-use-guide/](https://ai.meta.com/llama/responsible-use-guide)
|
AntonBOOM/output
|
AntonBOOM
| 2025-09-05T21:32:21Z | 0 | 0 |
transformers
|
[
"transformers",
"tensorboard",
"safetensors",
"generated_from_trainer",
"trl",
"sft",
"base_model:google/gemma-3-270m-it",
"base_model:finetune:google/gemma-3-270m-it",
"endpoints_compatible",
"region:us"
] | null | 2025-09-05T13:43:14Z |
---
base_model: google/gemma-3-270m-it
library_name: transformers
model_name: output
tags:
- generated_from_trainer
- trl
- sft
licence: license
---
# Model Card for output
This model is a fine-tuned version of [google/gemma-3-270m-it](https://huggingface.co/google/gemma-3-270m-it).
It has been trained using [TRL](https://github.com/huggingface/trl).
## Quick start
```python
from transformers import pipeline
question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
generator = pipeline("text-generation", model="AntonBOOM/output", device="cuda")
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
print(output["generated_text"])
```
## Training procedure
This model was trained with SFT.
### Framework versions
- TRL: 0.22.2
- Transformers: 4.56.0
- Pytorch: 2.8.0
- Datasets: 4.0.0
- Tokenizers: 0.22.0
## Citations
Cite TRL as:
```bibtex
@misc{vonwerra2022trl,
title = {{TRL: Transformer Reinforcement Learning}},
author = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallou{\'e}dec},
year = 2020,
journal = {GitHub repository},
publisher = {GitHub},
howpublished = {\url{https://github.com/huggingface/trl}}
}
```
|
Stasonelison/blockassist-bc-howling_powerful_aardvark_1757107862
|
Stasonelison
| 2025-09-05T21:31:58Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"howling powerful aardvark",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:31:48Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- howling powerful aardvark
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
Miracle-man/blockassist-bc-singing_lithe_koala_1757106050
|
Miracle-man
| 2025-09-05T21:31:48Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"singing lithe koala",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:31:44Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- singing lithe koala
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
forkkyty/blockassist-bc-skilled_omnivorous_elephant_1757107828
|
forkkyty
| 2025-09-05T21:30:45Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"skilled omnivorous elephant",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:30:29Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- skilled omnivorous elephant
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
taropok22/blockassist-bc-nasty_webbed_mouse_1757107784
|
taropok22
| 2025-09-05T21:30:38Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"nasty webbed mouse",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:30:18Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- nasty webbed mouse
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
andrewwentzel-epsilon/ttp-llama-Q8_0-GGUF
|
andrewwentzel-epsilon
| 2025-09-05T21:30:28Z | 0 | 0 |
transformers
|
[
"transformers",
"gguf",
"trl",
"sft",
"llama-cpp",
"gguf-my-repo",
"base_model:andrewwentzel-epsilon/ttp-llama",
"base_model:quantized:andrewwentzel-epsilon/ttp-llama",
"endpoints_compatible",
"region:us"
] | null | 2025-09-05T21:30:19Z |
---
library_name: transformers
tags:
- trl
- sft
- llama-cpp
- gguf-my-repo
base_model: andrewwentzel-epsilon/ttp-llama
---
# andrewwentzel-epsilon/ttp-llama-Q8_0-GGUF
This model was converted to GGUF format from [`andrewwentzel-epsilon/ttp-llama`](https://huggingface.co/andrewwentzel-epsilon/ttp-llama) using llama.cpp via the ggml.ai's [GGUF-my-repo](https://huggingface.co/spaces/ggml-org/gguf-my-repo) space.
Refer to the [original model card](https://huggingface.co/andrewwentzel-epsilon/ttp-llama) for more details on the model.
## Use with llama.cpp
Install llama.cpp through brew (works on Mac and Linux)
```bash
brew install llama.cpp
```
Invoke the llama.cpp server or the CLI.
### CLI:
```bash
llama-cli --hf-repo andrewwentzel-epsilon/ttp-llama-Q8_0-GGUF --hf-file ttp-llama-q8_0.gguf -p "The meaning to life and the universe is"
```
### Server:
```bash
llama-server --hf-repo andrewwentzel-epsilon/ttp-llama-Q8_0-GGUF --hf-file ttp-llama-q8_0.gguf -c 2048
```
Note: You can also use this checkpoint directly through the [usage steps](https://github.com/ggerganov/llama.cpp?tab=readme-ov-file#usage) listed in the Llama.cpp repo as well.
Step 1: Clone llama.cpp from GitHub.
```
git clone https://github.com/ggerganov/llama.cpp
```
Step 2: Move into the llama.cpp folder and build it with `LLAMA_CURL=1` flag along with other hardware-specific flags (for ex: LLAMA_CUDA=1 for Nvidia GPUs on Linux).
```
cd llama.cpp && LLAMA_CURL=1 make
```
Step 3: Run inference through the main binary.
```
./llama-cli --hf-repo andrewwentzel-epsilon/ttp-llama-Q8_0-GGUF --hf-file ttp-llama-q8_0.gguf -p "The meaning to life and the universe is"
```
or
```
./llama-server --hf-repo andrewwentzel-epsilon/ttp-llama-Q8_0-GGUF --hf-file ttp-llama-q8_0.gguf -c 2048
```
|
Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-IQ4_XS_R8-SPECIAL_SPLIT
|
Thireus
| 2025-09-05T21:30:12Z | 0 | 0 | null |
[
"gguf",
"arxiv:2505.23786",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-07-22T07:12:03Z |
---
license: mit
---
# DeepSeek-TNG-R1T2-Chimera
## 🤔 What is this [HuggingFace repository](https://huggingface.co/Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_SPLIT/) about?
This repository provides **GGUF-quantized tensors** for the DeepSeek-TNG-R1T2-Chimera model (official repo: https://huggingface.co/tngtech/DeepSeek-TNG-R1T2-Chimera). These GGUF shards are designed to be used with **Thireus’ GGUF Tool Suite** (https://gguf.thireus.com), a collection of tools that automatically finds the perplexity-optimal mix of quantizations for any given VRAM and RAM target. With the Tool Suite, you can generate and download custom quantization “recipes” effortlessly.
- 📖 Read more: https://github.com/Thireus/GGUF-Tool-Suite
- 🔍 Example quant mixes: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
- 🛠️ Create your own recipe: https://colab.research.google.com/github/Thireus/GGUF-Tool-Suite/blob/main/quant_recipe_pipeline.ipynb
- 📂 Browse available quant shards: https://huggingface.co/Thireus/collections
*tl;dr: Expand the details section below*
<details>
```
cd ~
# Make sure to install all ik_llama.cpp compilation dependencies...
apt install python3-dev python3-pip python3-venv python3-wheel python3-setuptools git acl netcat-openbsd cmake # pipx
# Obtain ik_llama's Thireus version - Windows builds available at https://github.com/Thireus/ik_llama.cpp/releases
git clone https://github.com/Thireus/ik_llama.cpp
cd ik_llama.cpp
git pull
# Build ik_llama.cpp
cmake -B build -DGGML_AVX=ON -DGGML_AVX2=ON -DLLAMA_CURL=OFF -DGGML_MAX_CONTEXTS=2048
cmake --build build --config Release -j16
cd ..
# Obtain Thireus' GGUF-Tool-Suite
git clone https://github.com/Thireus/GGUF-Tool-Suite
# Download model quant mix from recipe file:
cd GGUF-Tool-Suite
rm -f download.conf # Make sure to copy the relevant download.conf for the model before running quant_assign.py
cp -f models/DeepSeek-TNG-R1T2-Chimera/download.conf . # Use the download.conf of the chosen model
mkdir -p kitchen && cd kitchen
../quant_downloader.sh ../recipe_examples/ik_llama.cpp_recipes/DeepSeek-TNG-R1T2-Chimera.ROOT-3.0624bpw-3.3657ppl.238GB-GGUF_11GB-GPU_227GB-CPU.13549e6_1ac857a.recipe
# Launch ik_llama's llama-cli:
ulimit -n 9999 # Lifts "too many open files" limitation on Linux
~/ik_llama.cpp/build/bin/llama-cli \
-m DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_TENSOR-00001-of-01148.gguf \
-mla 3 -fa -amb 512 -fmoe -ctk f16 -c 4096 -ngl 99 \
-ot "blk\.(3|4|5|6)\.ffn_.*=CUDA0" \
-ot "blk\.(7|8|9|10)\.ffn_.*=CUDA1" \
-ot exps=CPU -b 2048 -ub 1024 --warmup-batch --no-mmap --threads 36 \
--main-gpu 0 \
-p '<｜begin▁of▁sentence｜><｜User｜>What is the solution of x+5=-2?<｜Assistant｜><think>\n'
```
</details>
---
## ❓ Why does this Tool Suite exist?
1. **Compatibility & Speed** – [unsloth](https://huggingface.co/unsloth)’s dynamic quants may not always work optimally with `ik_llama.cpp`.
2. **Custom Rig Fit** – No off-the-shelf GGUF model perfectly matched my VRAM/RAM setup, so I built a way to tailor models and leverage extra VRAM/RAM to reduce perplexity.
3. **Automated PPL-Optimal Quantization** – To my knowledge, there was no flexible, automated method to minimize perplexity for any bits-per-weight (bpw) target—so I created one with excellent results!
---
## 📊 How does it compare to other GGUFs?
Here’s how DeepSeek-R1-0528 quantized with **Thireus’ GGUF Tool Suite** stacks up against other quantizers (lower perplexity = better at equal or lower bpw):
> _Note: The `recipe_examples` files illustrate good recipes. The Tool Suite computes the optimal ppl/bpw curve for you — just specify your target RAM, VRAM, and quant types, and `quant_assign.py` finds the best mix._
More perplexity/bpw graphs for other supported models: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/ppl_graphs
*All PPL values are computed with the parameters `-ctk f16 -c 512 -b 4096 -ub 4096`. Changing any of these parameters will alter the PPL. In particular, reducing `-b 4096 -ub 4096` increases the PPL, while increasing them decreases the PPL.*
---
## 🚀 How do I get started?
Check out the [GGUF Tool Suite README](https://github.com/Thireus/GGUF-Tool-Suite) — focus on these sections:
1. ⚠️ **Requirements** – Which `ik_llama.cpp` (or `llama.cpp`) version to use and how to compile.
- Windows binaries (no patching needed) at: https://github.com/Thireus/ik_llama.cpp/releases
2. 📥 **Download Model Shards** – Use `quant_downloader.sh` to fetch GGUF shards from any recipe.
- Recipe examples: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
3. 🧠 **Run a Downloaded Model** – Sample usage with `llama-cli`.
4. 🛠️ **Generate a Custom Recipe** – Produce recipes tailored to your rig for optimal perplexity.
---
## ✅ Supported Models
Supported models are listed under `models/` in the [Tool Suite Github repo](https://github.com/Thireus/GGUF-Tool-Suite/tree/main/models). Presence of `ppl_results.csv` indicates official support and compatibility with `quant_assign.py`.
---
## 🤷♂️ Will I release pre-cooked GGUF files?
No, because I believe in **tailored quantization** for each user’s hardware. If you prefer ready-made shards, you are welcome to merge them via `llama-gguf-split --merge`, or request someone to publish them.
Instead, I prefer to share examples of recipes so users can see exactly how they were produced (command included inside these recipe files) and tweak them for their own rigs. The `quant_downloader.sh` script handles automatic fetching and verification of each shard. Recipes provided by [Ubergarm](https://huggingface.co/ubergarm) on his model cards are also compatible with `quant_downloader.sh`.
Users who don’t trust the GGUF shards on HuggingFace can also quantize their own by passing recipe lines to `llama-quantize --custom-q` ([see example](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/models/DeepSeek-R1-0528/DeepSeek-R1-0528-THIREUS-ANY-SPECIAL.sh#L482-L486)). Run `llama-quantize --help` to list compatible quants for `quant_assign.py`. This approach is especially useful if you prefer `llama.cpp` over `ik_llama.cpp`.
---
## 📦 What’s in this repository?
- **00001 GGUF header shard** – Contains metadata (tokens, chat template, tensor count, etc.). This metadata can be explored directly from the HuggingFace web interface after clicking on that shard.
- **Tensor shards** – Each shard holds one tensor; see `tensors.map` for names, quant types, sizes, SHA-256 hash, shard IDs, etc.
- **GPG-signed files** – `tensors.map` and header shard are signed with the key in [trusted-keys.asc](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/trusted-keys.asc) for tamper detection.
- **Security note** – Some papers about various ways to attack GGUFs and LLMs are available online, such as https://arxiv.org/abs/2505.23786, and there are also more classic security exploits like CVE-2024-23496 and CVE-2024-25664 through CVE-2024-25668. Only use GGUFs from reputable, trusted authors—or alternatively self-quantize—to avoid potential exploits.
---
## 💡 Pro Tips
You can download the BF16 model version to quantize your own shards:
```
mkdir kitchen
echo '.*=bf16' > kitchen/bf16.recipe
cd kitchen
../quant_downloader.sh bf16.recipe
```
Enjoy optimized quantization! 🎉
|
bah63843/blockassist-bc-plump_fast_antelope_1757107708
|
bah63843
| 2025-09-05T21:29:23Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"plump fast antelope",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:29:15Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- plump fast antelope
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
hdong0/qwen2_dummy_lora
|
hdong0
| 2025-09-05T21:28:22Z | 0 | 0 |
transformers
|
[
"transformers",
"safetensors",
"qwen2",
"text-generation",
"conversational",
"custom_code",
"arxiv:1910.09700",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] |
text-generation
| 2025-09-05T20:55:00Z |
---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed]
|
abcorrea/mix-4k
|
abcorrea
| 2025-09-05T21:27:49Z | 0 | 0 |
transformers
|
[
"transformers",
"safetensors",
"qwen3",
"text-generation",
"text-generation-inference",
"unsloth",
"conversational",
"en",
"base_model:unsloth/Qwen3-1.7B",
"base_model:finetune:unsloth/Qwen3-1.7B",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] |
text-generation
| 2025-09-05T21:26:30Z |
---
base_model: unsloth/Qwen3-1.7B
tags:
- text-generation-inference
- transformers
- unsloth
- qwen3
license: apache-2.0
language:
- en
---
# Uploaded finetuned model
- **Developed by:** abcorrea
- **License:** apache-2.0
- **Finetuned from model :** unsloth/Qwen3-1.7B
This qwen3 model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|
Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-IQ4_NL-SPECIAL_SPLIT
|
Thireus
| 2025-09-05T21:27:44Z | 0 | 0 | null |
[
"gguf",
"arxiv:2505.23786",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-07-20T00:48:14Z |
---
license: mit
---
# DeepSeek-TNG-R1T2-Chimera
## 🤔 What is this [HuggingFace repository](https://huggingface.co/Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_SPLIT/) about?
This repository provides **GGUF-quantized tensors** for the DeepSeek-TNG-R1T2-Chimera model (official repo: https://huggingface.co/tngtech/DeepSeek-TNG-R1T2-Chimera). These GGUF shards are designed to be used with **Thireus’ GGUF Tool Suite** (https://gguf.thireus.com), a collection of tools that automatically finds the perplexity-optimal mix of quantizations for any given VRAM and RAM target. With the Tool Suite, you can generate and download custom quantization “recipes” effortlessly.
- 📖 Read more: https://github.com/Thireus/GGUF-Tool-Suite
- 🔍 Example quant mixes: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
- 🛠️ Create your own recipe: https://colab.research.google.com/github/Thireus/GGUF-Tool-Suite/blob/main/quant_recipe_pipeline.ipynb
- 📂 Browse available quant shards: https://huggingface.co/Thireus/collections
*tl;dr: Expand the details section below*
<details>
```
cd ~
# Make sure to install all ik_llama.cpp compilation dependencies...
apt install python3-dev python3-pip python3-venv python3-wheel python3-setuptools git acl netcat-openbsd cmake # pipx
# Obtain ik_llama's Thireus version - Windows builds available at https://github.com/Thireus/ik_llama.cpp/releases
git clone https://github.com/Thireus/ik_llama.cpp
cd ik_llama.cpp
git pull
# Build ik_llama.cpp
cmake -B build -DGGML_AVX=ON -DGGML_AVX2=ON -DLLAMA_CURL=OFF -DGGML_MAX_CONTEXTS=2048
cmake --build build --config Release -j16
cd ..
# Obtain Thireus' GGUF-Tool-Suite
git clone https://github.com/Thireus/GGUF-Tool-Suite
# Download model quant mix from recipe file:
cd GGUF-Tool-Suite
rm -f download.conf # Make sure to copy the relevant download.conf for the model before running quant_assign.py
cp -f models/DeepSeek-TNG-R1T2-Chimera/download.conf . # Use the download.conf of the chosen model
mkdir -p kitchen && cd kitchen
../quant_downloader.sh ../recipe_examples/ik_llama.cpp_recipes/DeepSeek-TNG-R1T2-Chimera.ROOT-3.0624bpw-3.3657ppl.238GB-GGUF_11GB-GPU_227GB-CPU.13549e6_1ac857a.recipe
# Launch ik_llama's llama-cli:
ulimit -n 9999 # Lifts "too many open files" limitation on Linux
~/ik_llama.cpp/build/bin/llama-cli \
-m DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_TENSOR-00001-of-01148.gguf \
-mla 3 -fa -amb 512 -fmoe -ctk f16 -c 4096 -ngl 99 \
-ot "blk\.(3|4|5|6)\.ffn_.*=CUDA0" \
-ot "blk\.(7|8|9|10)\.ffn_.*=CUDA1" \
-ot exps=CPU -b 2048 -ub 1024 --warmup-batch --no-mmap --threads 36 \
--main-gpu 0 \
-p '<｜begin▁of▁sentence｜><｜User｜>What is the solution of x+5=-2?<｜Assistant｜><think>\n'
```
</details>
---
## ❓ Why does this Tool Suite exist?
1. **Compatibility & Speed** – [unsloth](https://huggingface.co/unsloth)’s dynamic quants may not always work optimally with `ik_llama.cpp`.
2. **Custom Rig Fit** – No off-the-shelf GGUF model perfectly matched my VRAM/RAM setup, so I built a way to tailor models and leverage extra VRAM/RAM to reduce perplexity.
3. **Automated PPL-Optimal Quantization** – To my knowledge, there was no flexible, automated method to minimize perplexity for any bits-per-weight (bpw) target—so I created one with excellent results!
---
## 📊 How does it compare to other GGUFs?
Here’s how DeepSeek-R1-0528 quantized with **Thireus’ GGUF Tool Suite** stacks up against other quantizers (lower perplexity = better at equal or lower bpw):
> _Note: The `recipe_examples` files illustrate good recipes. The Tool Suite computes the optimal ppl/bpw curve for you — just specify your target RAM, VRAM, and quant types, and `quant_assign.py` finds the best mix._
More perplexity/bpw graphs for other supported models: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/ppl_graphs
*All PPL values are computed with the parameters `-ctk f16 -c 512 -b 4096 -ub 4096`. Changing any of these parameters will alter the PPL. In particular, reducing `-b 4096 -ub 4096` increases the PPL, while increasing them decreases the PPL.*
---
## 🚀 How do I get started?
Check out the [GGUF Tool Suite README](https://github.com/Thireus/GGUF-Tool-Suite) — focus on these sections:
1. ⚠️ **Requirements** – Which `ik_llama.cpp` (or `llama.cpp`) version to use and how to compile.
- Windows binaries (no patching needed) at: https://github.com/Thireus/ik_llama.cpp/releases
2. 📥 **Download Model Shards** – Use `quant_downloader.sh` to fetch GGUF shards from any recipe.
- Recipe examples: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
3. 🧠 **Run a Downloaded Model** – Sample usage with `llama-cli`.
4. 🛠️ **Generate a Custom Recipe** – Produce recipes tailored to your rig for optimal perplexity.
---
## ✅ Supported Models
Supported models are listed under `models/` in the [Tool Suite Github repo](https://github.com/Thireus/GGUF-Tool-Suite/tree/main/models). Presence of `ppl_results.csv` indicates official support and compatibility with `quant_assign.py`.
---
## 🤷♂️ Will I release pre-cooked GGUF files?
No, because I believe in **tailored quantization** for each user’s hardware. If you prefer ready-made shards, you are welcome to merge them via `llama-gguf-split --merge`, or request someone to publish them.
Instead, I prefer to share examples of recipes so users can see exactly how they were produced (command included inside these recipe files) and tweak them for their own rigs. The `quant_downloader.sh` script handles automatic fetching and verification of each shard. Recipes provided by [Ubergarm](https://huggingface.co/ubergarm) on his model cards are also compatible with `quant_downloader.sh`.
Users who don’t trust the GGUF shards on HuggingFace can also quantize their own by passing recipe lines to `llama-quantize --custom-q` ([see example](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/models/DeepSeek-R1-0528/DeepSeek-R1-0528-THIREUS-ANY-SPECIAL.sh#L482-L486)). Run `llama-quantize --help` to list compatible quants for `quant_assign.py`. This approach is especially useful if you prefer `llama.cpp` over `ik_llama.cpp`.
---
## 📦 What’s in this repository?
- **00001 GGUF header shard** – Contains metadata (tokens, chat template, tensor count, etc.). This metadata can be explored directly from the HuggingFace web interface after clicking on that shard.
- **Tensor shards** – Each shard holds one tensor; see `tensors.map` for names, quant types, sizes, SHA-256 hash, shard IDs, etc.
- **GPG-signed files** – `tensors.map` and header shard are signed with the key in [trusted-keys.asc](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/trusted-keys.asc) for tamper detection.
- **Security note** – Some papers about various ways to attack GGUFs and LLMs are available online, such as https://arxiv.org/abs/2505.23786, and there are also more classic security exploits like CVE-2024-23496 and CVE-2024-25664 through CVE-2024-25668. Only use GGUFs from reputable, trusted authors—or alternatively self-quantize—to avoid potential exploits.
---
## 💡 Pro Tips
You can download the BF16 model version to quantize your own shards:
```
mkdir kitchen
echo '.*=bf16' > kitchen/bf16.recipe
cd kitchen
../quant_downloader.sh bf16.recipe
```
Enjoy optimized quantization! 🎉
|
rocktanmay2012/blockassist-bc-bold_placid_barracuda_1757107564
|
rocktanmay2012
| 2025-09-05T21:26:45Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"bold placid barracuda",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:26:33Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- bold placid barracuda
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-IQ4_KS_R4-SPECIAL_SPLIT
|
Thireus
| 2025-09-05T21:25:20Z | 91 | 0 | null |
[
"gguf",
"arxiv:2505.23786",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-07-20T07:29:29Z |
---
license: mit
---
# DeepSeek-TNG-R1T2-Chimera
## 🤔 What is this [HuggingFace repository](https://huggingface.co/Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_SPLIT/) about?
This repository provides **GGUF-quantized tensors** for the DeepSeek-TNG-R1T2-Chimera model (official repo: https://huggingface.co/tngtech/DeepSeek-TNG-R1T2-Chimera). These GGUF shards are designed to be used with **Thireus’ GGUF Tool Suite** (https://gguf.thireus.com), a collection of tools that automatically finds the perplexity-optimal mix of quantizations for any given VRAM and RAM target. With the Tool Suite, you can generate and download custom quantization “recipes” effortlessly.
- 📖 Read more: https://github.com/Thireus/GGUF-Tool-Suite
- 🔍 Example quant mixes: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
- 🛠️ Create your own recipe: https://colab.research.google.com/github/Thireus/GGUF-Tool-Suite/blob/main/quant_recipe_pipeline.ipynb
- 📂 Browse available quant shards: https://huggingface.co/Thireus/collections
*tl;dr: Expand the details section below*
<details>
```
cd ~
# Make sure to install all ik_llama.cpp compilation dependencies...
apt install python3-dev python3-pip python3-venv python3-wheel python3-setuptools git acl netcat-openbsd cmake # pipx
# Obtain ik_llama's Thireus version - Windows builds available at https://github.com/Thireus/ik_llama.cpp/releases
git clone https://github.com/Thireus/ik_llama.cpp
cd ik_llama.cpp
git pull
# Build ik_llama.cpp
cmake -B build -DGGML_AVX=ON -DGGML_AVX2=ON -DLLAMA_CURL=OFF -DGGML_MAX_CONTEXTS=2048
cmake --build build --config Release -j16
cd ..
# Obtain Thireus' GGUF-Tool-Suite
git clone https://github.com/Thireus/GGUF-Tool-Suite
# Download model quant mix from recipe file:
cd GGUF-Tool-Suite
rm -f download.conf # Make sure to copy the relevant download.conf for the model before running quant_assign.py
cp -f models/DeepSeek-TNG-R1T2-Chimera/download.conf . # Use the download.conf of the chosen model
mkdir -p kitchen && cd kitchen
../quant_downloader.sh ../recipe_examples/ik_llama.cpp_recipes/DeepSeek-TNG-R1T2-Chimera.ROOT-3.0624bpw-3.3657ppl.238GB-GGUF_11GB-GPU_227GB-CPU.13549e6_1ac857a.recipe
# Launch ik_llama's llama-cli:
ulimit -n 9999 # Lifts "too many open files" limitation on Linux
~/ik_llama.cpp/build/bin/llama-cli \
-m DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_TENSOR-00001-of-01148.gguf \
-mla 3 -fa -amb 512 -fmoe -ctk f16 -c 4096 -ngl 99 \
-ot "blk\.(3|4|5|6)\.ffn_.*=CUDA0" \
-ot "blk\.(7|8|9|10)\.ffn_.*=CUDA1" \
-ot exps=CPU -b 2048 -ub 1024 --warmup-batch --no-mmap --threads 36 \
--main-gpu 0 \
-p '<｜begin▁of▁sentence｜><｜User｜>What is the solution of x+5=-2?<｜Assistant｜><think>\n'
```
</details>
---
## ❓ Why does this Tool Suite exist?
1. **Compatibility & Speed** – [unsloth](https://huggingface.co/unsloth)’s dynamic quants may not always work optimally with `ik_llama.cpp`.
2. **Custom Rig Fit** – No off-the-shelf GGUF model perfectly matched my VRAM/RAM setup, so I built a way to tailor models and leverage extra VRAM/RAM to reduce perplexity.
3. **Automated PPL-Optimal Quantization** – To my knowledge, there was no flexible, automated method to minimize perplexity for any bits-per-weight (bpw) target—so I created one with excellent results!
---
## 📊 How does it compare to other GGUFs?
Here’s how DeepSeek-R1-0528 quantized with **Thireus’ GGUF Tool Suite** stacks up against other quantizers (lower perplexity = better at equal or lower bpw):
> _Note: The `recipe_examples` files illustrate good recipes. The Tool Suite computes the optimal ppl/bpw curve for you — just specify your target RAM, VRAM, and quant types, and `quant_assign.py` finds the best mix._
More perplexity/bpw graphs for other supported models: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/ppl_graphs
*All PPL values are computed with the parameters `-ctk f16 -c 512 -b 4096 -ub 4096`. Changing any of these parameters will alter the PPL. In particular, reducing `-b 4096 -ub 4096` increases the PPL, while increasing them decreases the PPL.*
---
## 🚀 How do I get started?
Check out the [GGUF Tool Suite README](https://github.com/Thireus/GGUF-Tool-Suite) — focus on these sections:
1. ⚠️ **Requirements** – Which `ik_llama.cpp` (or `llama.cpp`) version to use and how to compile.
- Windows binaries (no patching needed) at: https://github.com/Thireus/ik_llama.cpp/releases
2. 📥 **Download Model Shards** – Use `quant_downloader.sh` to fetch GGUF shards from any recipe.
- Recipe examples: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
3. 🧠 **Run a Downloaded Model** – Sample usage with `llama-cli`.
4. 🛠️ **Generate a Custom Recipe** – Produce recipes tailored to your rig for optimal perplexity.
---
## ✅ Supported Models
Supported models are listed under `models/` in the [Tool Suite Github repo](https://github.com/Thireus/GGUF-Tool-Suite/tree/main/models). Presence of `ppl_results.csv` indicates official support and compatibility with `quant_assign.py`.
---
## 🤷♂️ Will I release pre-cooked GGUF files?
No, because I believe in **tailored quantization** for each user’s hardware. If you prefer ready-made shards, you are welcome to merge them via `llama-gguf-split --merge`, or request someone to publish them.
Instead, I prefer to share examples of recipes so users can see exactly how they were produced (command included inside these recipe files) and tweak them for their own rigs. The `quant_downloader.sh` script handles automatic fetching and verification of each shard. Recipes provided by [Ubergarm](https://huggingface.co/ubergarm) on his model cards are also compatible with `quant_downloader.sh`.
Users who don’t trust the GGUF shards on HuggingFace can also quantize their own by passing recipe lines to `llama-quantize --custom-q` ([see example](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/models/DeepSeek-R1-0528/DeepSeek-R1-0528-THIREUS-ANY-SPECIAL.sh#L482-L486)). Run `llama-quantize --help` to list compatible quants for `quant_assign.py`. This approach is especially useful if you prefer `llama.cpp` over `ik_llama.cpp`.
---
## 📦 What’s in this repository?
- **00001 GGUF header shard** – Contains metadata (tokens, chat template, tensor count, etc.). This metadata can be explored directly from the HuggingFace web interface after clicking on that shard.
- **Tensor shards** – Each shard holds one tensor; see `tensors.map` for names, quant types, sizes, SHA-256 hash, shard IDs, etc.
- **GPG-signed files** – `tensors.map` and header shard are signed with the key in [trusted-keys.asc](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/trusted-keys.asc) for tamper detection.
- **Security note** – Some papers about various ways to attack GGUFs and LLMs are available online, such as https://arxiv.org/abs/2505.23786, and there are also more classic security exploits like CVE-2024-23496 and CVE-2024-25664 through CVE-2024-25668. Only use GGUFs from reputable, trusted authors—or alternatively self-quantize—to avoid potential exploits.
---
## 💡 Pro Tips
You can download the BF16 model version to quantize your own shards:
```
mkdir kitchen
echo '.*=bf16' > kitchen/bf16.recipe
cd kitchen
../quant_downloader.sh bf16.recipe
```
Enjoy optimized quantization! 🎉
|
Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-IQ4_K_R4-SPECIAL_SPLIT
|
Thireus
| 2025-09-05T21:24:10Z | 0 | 0 | null |
[
"gguf",
"arxiv:2505.23786",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-07-21T20:38:33Z |
---
license: mit
---
# DeepSeek-TNG-R1T2-Chimera
## 🤔 What is this [HuggingFace repository](https://huggingface.co/Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_SPLIT/) about?
This repository provides **GGUF-quantized tensors** for the DeepSeek-TNG-R1T2-Chimera model (official repo: https://huggingface.co/tngtech/DeepSeek-TNG-R1T2-Chimera). These GGUF shards are designed to be used with **Thireus’ GGUF Tool Suite** (https://gguf.thireus.com), a collection of tools that automatically finds the perplexity-optimal mix of quantizations for any given VRAM and RAM target. With the Tool Suite, you can generate and download custom quantization “recipes” effortlessly.
- 📖 Read more: https://github.com/Thireus/GGUF-Tool-Suite
- 🔍 Example quant mixes: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
- 🛠️ Create your own recipe: https://colab.research.google.com/github/Thireus/GGUF-Tool-Suite/blob/main/quant_recipe_pipeline.ipynb
- 📂 Browse available quant shards: https://huggingface.co/Thireus/collections
*tl;dr: Expand the details section below*
<details>
```
cd ~
# Make sure to install all ik_llama.cpp compilation dependencies...
apt install python3-dev python3-pip python3-venv python3-wheel python3-setuptools git acl netcat-openbsd cmake # pipx
# Obtain ik_llama's Thireus version - Windows builds available at https://github.com/Thireus/ik_llama.cpp/releases
git clone https://github.com/Thireus/ik_llama.cpp
cd ik_llama.cpp
git pull
# Build ik_llama.cpp
cmake -B build -DGGML_AVX=ON -DGGML_AVX2=ON -DLLAMA_CURL=OFF -DGGML_MAX_CONTEXTS=2048
cmake --build build --config Release -j16
cd ..
# Obtain Thireus' GGUF-Tool-Suite
git clone https://github.com/Thireus/GGUF-Tool-Suite
# Download model quant mix from recipe file:
cd GGUF-Tool-Suite
rm -f download.conf # Make sure to copy the relevant download.conf for the model before running quant_assign.py
cp -f models/DeepSeek-TNG-R1T2-Chimera/download.conf . # Use the download.conf of the chosen model
mkdir -p kitchen && cd kitchen
../quant_downloader.sh ../recipe_examples/ik_llama.cpp_recipes/DeepSeek-TNG-R1T2-Chimera.ROOT-3.0624bpw-3.3657ppl.238GB-GGUF_11GB-GPU_227GB-CPU.13549e6_1ac857a.recipe
# Launch ik_llama's llama-cli:
ulimit -n 9999 # Lifts "too many open files" limitation on Linux
~/ik_llama.cpp/build/bin/llama-cli \
-m DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_TENSOR-00001-of-01148.gguf \
-mla 3 -fa -amb 512 -fmoe -ctk f16 -c 4096 -ngl 99 \
-ot "blk\.(3|4|5|6)\.ffn_.*=CUDA0" \
-ot "blk\.(7|8|9|10)\.ffn_.*=CUDA1" \
-ot exps=CPU -b 2048 -ub 1024 --warmup-batch --no-mmap --threads 36 \
--main-gpu 0 \
-p '<｜begin▁of▁sentence｜><｜User｜>What is the solution of x+5=-2?<｜Assistant｜><think>\n'
```
</details>
---
## ❓ Why does this Tool Suite exist?
1. **Compatibility & Speed** – [unsloth](https://huggingface.co/unsloth)’s dynamic quants may not always work optimally with `ik_llama.cpp`.
2. **Custom Rig Fit** – No off-the-shelf GGUF model perfectly matched my VRAM/RAM setup, so I built a way to tailor models and leverage extra VRAM/RAM to reduce perplexity.
3. **Automated PPL-Optimal Quantization** – To my knowledge, there was no flexible, automated method to minimize perplexity for any bits-per-weight (bpw) target—so I created one with excellent results!
---
## 📊 How does it compare to other GGUFs?
Here’s how DeepSeek-R1-0528 quantized with **Thireus’ GGUF Tool Suite** stacks up against other quantizers (lower perplexity = better at equal or lower bpw):
> _Note: The `recipe_examples` files illustrate good recipes. The Tool Suite computes the optimal ppl/bpw curve for you — just specify your target RAM, VRAM, and quant types, and `quant_assign.py` finds the best mix._
More perplexity/bpw graphs for other supported models: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/ppl_graphs
*All PPL values are computed with the parameters `-ctk f16 -c 512 -b 4096 -ub 4096`. Changing any of these parameters will alter the PPL. In particular, reducing `-b 4096 -ub 4096` increases the PPL, while increasing them decreases the PPL.*
---
## 🚀 How do I get started?
Check out the [GGUF Tool Suite README](https://github.com/Thireus/GGUF-Tool-Suite) — focus on these sections:
1. ⚠️ **Requirements** – Which `ik_llama.cpp` (or `llama.cpp`) version to use and how to compile.
- Windows binaries (no patching needed) at: https://github.com/Thireus/ik_llama.cpp/releases
2. 📥 **Download Model Shards** – Use `quant_downloader.sh` to fetch GGUF shards from any recipe.
- Recipe examples: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
3. 🧠 **Run a Downloaded Model** – Sample usage with `llama-cli`.
4. 🛠️ **Generate a Custom Recipe** – Produce recipes tailored to your rig for optimal perplexity.
---
## ✅ Supported Models
Supported models are listed under `models/` in the [Tool Suite Github repo](https://github.com/Thireus/GGUF-Tool-Suite/tree/main/models). Presence of `ppl_results.csv` indicates official support and compatibility with `quant_assign.py`.
---
## 🤷♂️ Will I release pre-cooked GGUF files?
No, because I believe in **tailored quantization** for each user’s hardware. If you prefer ready-made shards, you are welcome to merge them via `llama-gguf-split --merge`, or request someone to publish them.
Instead, I prefer to share examples of recipes so users can see exactly how they were produced (command included inside these recipe files) and tweak them for their own rigs. The `quant_downloader.sh` script handles automatic fetching and verification of each shard. Recipes provided by [Ubergarm](https://huggingface.co/ubergarm) on his model cards are also compatible with `quant_downloader.sh`.
Users who don’t trust the GGUF shards on HuggingFace can also quantize their own by passing recipe lines to `llama-quantize --custom-q` ([see example](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/models/DeepSeek-R1-0528/DeepSeek-R1-0528-THIREUS-ANY-SPECIAL.sh#L482-L486)). Run `llama-quantize --help` to list compatible quants for `quant_assign.py`. This approach is especially useful if you prefer `llama.cpp` over `ik_llama.cpp`.
---
## 📦 What’s in this repository?
- **00001 GGUF header shard** – Contains metadata (tokens, chat template, tensor count, etc.). This metadata can be explored directly from the HuggingFace web interface after clicking on that shard.
- **Tensor shards** – Each shard holds one tensor; see `tensors.map` for names, quant types, sizes, SHA-256 hash, shard IDs, etc.
- **GPG-signed files** – `tensors.map` and header shard are signed with the key in [trusted-keys.asc](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/trusted-keys.asc) for tamper detection.
- **Security note** – Some papers about various ways to attack GGUFs and LLMs are available online, such as https://arxiv.org/abs/2505.23786, and there are also more classic security exploits like CVE-2024-23496 and CVE-2024-25664 through CVE-2024-25668. Only use GGUFs from reputable, trusted authors—or alternatively self-quantize—to avoid potential exploits.
---
## 💡 Pro Tips
You can download the BF16 model version to quantize your own shards:
```
mkdir kitchen
echo '.*=bf16' > kitchen/bf16.recipe
cd kitchen
../quant_downloader.sh bf16.recipe
```
Enjoy optimized quantization! 🎉
|
vwzyrraz7l/blockassist-bc-tall_hunting_vulture_1757105964
|
vwzyrraz7l
| 2025-09-05T21:23:36Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"tall hunting vulture",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:23:32Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- tall hunting vulture
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
matteoangeloni/EduDolphin
|
matteoangeloni
| 2025-09-05T21:23:14Z | 0 | 0 |
transformers
|
[
"transformers",
"safetensors",
"llama",
"text-generation",
"text-generation-inference",
"unsloth",
"educational",
"en",
"it",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] |
text-generation
| 2025-09-05T20:47:55Z |
---
base_model: unsloth/meta-llama-3.1-8b-unsloth-bnb-4bit
tags:
- text-generation-inference
- transformers
- unsloth
- llama
- educational
license: apache-2.0
language:
- en
- it
---
# EduDolphin 🐬📚
<p align="center">
<img src="./edudolphin_logo.png" alt="EduDolphin Logo" width="240"/>
</p>
**A fine‑tuned Llama 3.1 8B model specialized for learning analytics and academic insights.**
> TL;DR — EduDolphin analyzes educational datasets to surface patterns in student performance, engagement (VLE), demographics, and assessment design. Trained on carefully crafted prompts derived from OULAD. Use the **Alpaca‑style prompt template** below.
---
## Model Summary
* **Developer**: Matteo Angeloni ([@matteoangeloni](https://huggingface.co/matteoangeloni))
* **Base model**: `meta-llama/Meta-Llama-3.1-8B`
* **Method**: LoRA fine‑tuning with **Unsloth** + **TRL**
* **Primary artifact**: merged **FP16** (safetensors)
* **Other artifacts**: LoRA adapters; optional 4‑bit merged (env‑sensitive)
* **Languages**: English
* **Domain**: Educational Data / Learning Analytics
* **License**: **Llama 3** — access requires accepting Meta’s license on the Hub (gated)
## Intended Uses
### Primary
* **Learning Analytics**: detect performance patterns, retention risks, intervention windows.
* **Assessment Analytics**: reason over assessment types (TMA/CMA/exams), timing, grade distributions.
* **Demographics & Equity**: surface correlations and disparities in outcomes.
* **VLE Behavior**: interpret clickstream/engagement sequences across weeks and materials.
* **Academic Planning**: support course design decisions with evidence‑oriented insights.
### Limitations / Out‑of‑Scope
* High‑stakes **automated decision‑making** without human review.
* Any **non‑anonymized** student data processing (you must anonymize upstream).
* General domain tasks unrelated to education (the model is domain‑biased).
## Prompting Format (Alpaca)
Use this template for best results:
```
Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
### Instruction:
{instruction}
### Input:
{input}
### Response:
```
### Minimal Example
```python
from transformers import AutoModelForCausalLM, AutoTokenizer, GenerationConfig
import torch
MODEL = "matteoangeloni/EduDolphin"
model = AutoModelForCausalLM.from_pretrained(
MODEL,
torch_dtype=torch.float16,
device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(MODEL)
prompt = (
"Below is an instruction that describes a task, paired with an input that provides further context. "
"Write a response that appropriately completes the request.
"
"### Instruction:
"
"Task: Assessment Performance Analysis for Module AAA (Category: Learning Analytics)
"
"### Input:
"
"Analyze the assessment performance data for module AAA. We have 2,847 total submissions "
"with an average score of 67.3% and a pass rate of 71.2%. What insights can you derive?
"
"### Response:
"
)
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
# (optional) override default generation settings
model.generation_config = GenerationConfig(max_new_tokens=256, temperature=0.7, top_p=0.9)
outputs = model.generate(**inputs)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```
## Files & Variants
| Artifact | Purpose | Notes |
| ----------------------------- | ------------------------------------------------ | -------------------------------------------------------------- |
| **FP16 merged (default)** | Ready‑to‑use full model with LoRA merged | Recommended for most users; broad backend support |
| **LoRA adapters** | Combine with base `meta-llama/Meta-Llama-3.1-8B` | Smaller download; flexible for further finetuning |
| **4‑bit merged** *(optional)* | Lower footprint | Requires `bitsandbytes`; not all runtimes (e.g., some TGI/TEI) |
> Always distribute **tokenizer** and a **generation\_config.json** alongside weights to avoid inference mismatches.
## Training Data
**Source**: Open University Learning Analytics Dataset (**OULAD**)
**Underlying tables (original OULAD):**
* \~173,912 student assessment records
* \~10,655,280 VLE interaction logs
* \~32,593 student demographic profiles
* 6,364 learning material records
* 206 assessment configurations
**Prompt dataset (derived from OULAD):** **6,215 examples** total
* Train: **5,593**
* Validation: **622**
**Categories covered (examples):**
1. Individual Material Analytics (4,781)
2. Weekly Engagement Analytics (878)
3. Complex Demographic Analytics (353)
4. Granular Performance Analytics (64)
5. Submission Timing Analytics (38)
6. Click Behavior Analytics (35)
7. Learning Journey Analytics (33)
8. Registration Timing Analytics (33)
> Notes: Data were anonymized/aggregated for prompt construction. No raw personal identifiers are included.
## Training Procedure
* **Framework**: **Unsloth** + **Hugging Face TRL**
* **Base Model**: Llama 3.1 8B
* **Finetuning**: **LoRA**
* **Epochs**: 2
* **Batch size (per device)**: 8
* **Gradient Accumulation**: 8
* **Learning Rate**: 2e-5
* **Max Seq Len**: 1024
* **Optimizer**: AdamW (8‑bit)
* **Speed‑ups**: Unsloth (\~2× faster)
### Export & Publishing
* Publish **FP16 merged** as the primary artifact.
* Also publish **LoRA adapters** for flexibility.
* 4‑bit merged is optional and environment‑sensitive.
* Include `tokenizer/` and `generation_config.json` in each artifact folder.
## Evaluation (Current Status)
No standardized benchmark is reported yet. Internal checks focused on:
* Faithfulness of schema‑aware reasoning over OULAD‑like contexts
* Consistency of recommendations given aggregate statistics
* Stability under temperature variations (0.2–0.9)
> Community PRs with rigorous evaluation suites are welcome.
## Ethical Considerations
* **Privacy**: Use only anonymized/aggregated student data. Comply with GDPR/institutional policies.
* **Bias & Fairness**: OULAD reflects a specific context; validate insights locally before action.
* **Human Oversight**: Treat outputs as decision support, not decisions.
* **Transparency**: Disclose AI assistance in analyses/reports.
## Security & Access
* **Do NOT hard‑code tokens**. Use env vars (e.g., `HF_TOKEN`). Revoke any exposed token immediately.
* **License**: Llama 3. Users must accept Meta’s license on the Hub. Consider enabling **gated access**.
## How to Cite
```bibtex
@misc{angeloni2024edudolphin,
title = {EduDolphin: A Fine-tuned Language Model for Educational Data Analysis},
author = {Matteo Angeloni},
year = {2024},
howpublished = {Hugging Face Model Hub},
url = {https://huggingface.co/matteoangeloni/EduDolphin}
}
```
## Acknowledgments
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
Thanks to **Unsloth** for efficient fine‑tuning tooling, **Hugging Face TRL** for training utilities, and **OULAD** for the public dataset.
---
### Quick Setup
```bash
pip install --upgrade transformers accelerate
# Optional (for 4-bit merges)
pip install bitsandbytes
```
|
Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-IQ4_K-SPECIAL_SPLIT
|
Thireus
| 2025-09-05T21:22:59Z | 0 | 0 | null |
[
"gguf",
"arxiv:2505.23786",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-07-20T07:29:51Z |
---
license: mit
---
# DeepSeek-TNG-R1T2-Chimera
## 🤔 What is this [HuggingFace repository](https://huggingface.co/Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_SPLIT/) about?
This repository provides **GGUF-quantized tensors** for the DeepSeek-TNG-R1T2-Chimera model (official repo: https://huggingface.co/tngtech/DeepSeek-TNG-R1T2-Chimera). These GGUF shards are designed to be used with **Thireus’ GGUF Tool Suite** (https://gguf.thireus.com), a collection of tools that automatically finds the perplexity-optimal mix of quantizations for any given VRAM and RAM target. With the Tool Suite, you can generate and download custom quantization “recipes” effortlessly.
- 📖 Read more: https://github.com/Thireus/GGUF-Tool-Suite
- 🔍 Example quant mixes: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
- 🛠️ Create your own recipe: https://colab.research.google.com/github/Thireus/GGUF-Tool-Suite/blob/main/quant_recipe_pipeline.ipynb
- 📂 Browse available quant shards: https://huggingface.co/Thireus/collections
*tl;dr: Expand the details section below*
<details>
```
cd ~
# Make sure to install all ik_llama.cpp compilation dependencies...
apt install python3-dev python3-pip python3-venv python3-wheel python3-setuptools git acl netcat-openbsd cmake # pipx
# Obtain ik_llama's Thireus version - Windows builds available at https://github.com/Thireus/ik_llama.cpp/releases
git clone https://github.com/Thireus/ik_llama.cpp
cd ik_llama.cpp
git pull
# Build ik_llama.cpp
cmake -B build -DGGML_AVX=ON -DGGML_AVX2=ON -DLLAMA_CURL=OFF -DGGML_MAX_CONTEXTS=2048
cmake --build build --config Release -j16
cd ..
# Obtain Thireus' GGUF-Tool-Suite
git clone https://github.com/Thireus/GGUF-Tool-Suite
# Download model quant mix from recipe file:
cd GGUF-Tool-Suite
rm -f download.conf # Make sure to copy the relevant download.conf for the model before running quant_assign.py
cp -f models/DeepSeek-TNG-R1T2-Chimera/download.conf . # Use the download.conf of the chosen model
mkdir -p kitchen && cd kitchen
../quant_downloader.sh ../recipe_examples/ik_llama.cpp_recipes/DeepSeek-TNG-R1T2-Chimera.ROOT-3.0624bpw-3.3657ppl.238GB-GGUF_11GB-GPU_227GB-CPU.13549e6_1ac857a.recipe
# Launch ik_llama's llama-cli:
ulimit -n 9999 # Lifts "too many open files" limitation on Linux
~/ik_llama.cpp/build/bin/llama-cli \
-m DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_TENSOR-00001-of-01148.gguf \
-mla 3 -fa -amb 512 -fmoe -ctk f16 -c 4096 -ngl 99 \
-ot "blk\.(3|4|5|6)\.ffn_.*=CUDA0" \
-ot "blk\.(7|8|9|10)\.ffn_.*=CUDA1" \
-ot exps=CPU -b 2048 -ub 1024 --warmup-batch --no-mmap --threads 36 \
--main-gpu 0 \
-p '<｜begin▁of▁sentence｜><｜User｜>What is the solution of x+5=-2?<｜Assistant｜><think>\n'
```
</details>
---
## ❓ Why does this Tool Suite exist?
1. **Compatibility & Speed** – [unsloth](https://huggingface.co/unsloth)’s dynamic quants may not always work optimally with `ik_llama.cpp`.
2. **Custom Rig Fit** – No off-the-shelf GGUF model perfectly matched my VRAM/RAM setup, so I built a way to tailor models and leverage extra VRAM/RAM to reduce perplexity.
3. **Automated PPL-Optimal Quantization** – To my knowledge, there was no flexible, automated method to minimize perplexity for any bits-per-weight (bpw) target—so I created one with excellent results!
---
## 📊 How does it compare to other GGUFs?
Here’s how DeepSeek-R1-0528 quantized with **Thireus’ GGUF Tool Suite** stacks up against other quantizers (lower perplexity = better at equal or lower bpw):
> _Note: The `recipe_examples` files illustrate good recipes. The Tool Suite computes the optimal ppl/bpw curve for you — just specify your target RAM, VRAM, and quant types, and `quant_assign.py` finds the best mix._
More perplexity/bpw graphs for other supported models: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/ppl_graphs
*All PPL values are computed with the parameters `-ctk f16 -c 512 -b 4096 -ub 4096`. Changing any of these parameters will alter the PPL. In particular, reducing `-b 4096 -ub 4096` increases the PPL, while increasing them decreases the PPL.*
---
## 🚀 How do I get started?
Check out the [GGUF Tool Suite README](https://github.com/Thireus/GGUF-Tool-Suite) — focus on these sections:
1. ⚠️ **Requirements** – Which `ik_llama.cpp` (or `llama.cpp`) version to use and how to compile.
- Windows binaries (no patching needed) at: https://github.com/Thireus/ik_llama.cpp/releases
2. 📥 **Download Model Shards** – Use `quant_downloader.sh` to fetch GGUF shards from any recipe.
- Recipe examples: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
3. 🧠 **Run a Downloaded Model** – Sample usage with `llama-cli`.
4. 🛠️ **Generate a Custom Recipe** – Produce recipes tailored to your rig for optimal perplexity.
---
## ✅ Supported Models
Supported models are listed under `models/` in the [Tool Suite Github repo](https://github.com/Thireus/GGUF-Tool-Suite/tree/main/models). Presence of `ppl_results.csv` indicates official support and compatibility with `quant_assign.py`.
---
## 🤷♂️ Will I release pre-cooked GGUF files?
No, because I believe in **tailored quantization** for each user’s hardware. If you prefer ready-made shards, you are welcome to merge them via `llama-gguf-split --merge`, or request someone to publish them.
Instead, I prefer to share examples of recipes so users can see exactly how they were produced (command included inside these recipe files) and tweak them for their own rigs. The `quant_downloader.sh` script handles automatic fetching and verification of each shard. Recipes provided by [Ubergarm](https://huggingface.co/ubergarm) on his model cards are also compatible with `quant_downloader.sh`.
Users who don’t trust the GGUF shards on HuggingFace can also quantize their own by passing recipe lines to `llama-quantize --custom-q` ([see example](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/models/DeepSeek-R1-0528/DeepSeek-R1-0528-THIREUS-ANY-SPECIAL.sh#L482-L486)). Run `llama-quantize --help` to list compatible quants for `quant_assign.py`. This approach is especially useful if you prefer `llama.cpp` over `ik_llama.cpp`.
---
## 📦 What’s in this repository?
- **00001 GGUF header shard** – Contains metadata (tokens, chat template, tensor count, etc.). This metadata can be explored directly from the HuggingFace web interface after clicking on that shard.
- **Tensor shards** – Each shard holds one tensor; see `tensors.map` for names, quant types, sizes, SHA-256 hash, shard IDs, etc.
- **GPG-signed files** – `tensors.map` and header shard are signed with the key in [trusted-keys.asc](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/trusted-keys.asc) for tamper detection.
- **Security note** – Some papers about various ways to attack GGUFs and LLMs are available online, such as https://arxiv.org/abs/2505.23786, and there are also more classic security exploits like CVE-2024-23496 and CVE-2024-25664 through CVE-2024-25668. Only use GGUFs from reputable, trusted authors—or alternatively self-quantize—to avoid potential exploits.
---
## 💡 Pro Tips
You can download the BF16 model version to quantize your own shards:
```
mkdir kitchen
echo '.*=bf16' > kitchen/bf16.recipe
cd kitchen
../quant_downloader.sh bf16.recipe
```
Enjoy optimized quantization! 🎉
|
acidjp/blockassist-bc-pesty_extinct_prawn_1757104999
|
acidjp
| 2025-09-05T21:21:59Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"pesty extinct prawn",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:21:54Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- pesty extinct prawn
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
kavpro/blockassist-bc-tall_lively_caribou_1757107249
|
kavpro
| 2025-09-05T21:21:48Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"tall lively caribou",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:21:34Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- tall lively caribou
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
klmdr22/blockassist-bc-wild_loud_newt_1757107256
|
klmdr22
| 2025-09-05T21:21:38Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"wild loud newt",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:21:35Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- wild loud newt
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
tanvirahmedkhan/blockassist-bc-hardy_whiskered_mantis_1757107115
|
tanvirahmedkhan
| 2025-09-05T21:21:13Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"hardy whiskered mantis",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:20:49Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- hardy whiskered mantis
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
bah63843/blockassist-bc-plump_fast_antelope_1757107195
|
bah63843
| 2025-09-05T21:20:48Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"plump fast antelope",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:20:39Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- plump fast antelope
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-IQ3_S_R4-SPECIAL_SPLIT
|
Thireus
| 2025-09-05T21:20:34Z | 0 | 0 | null |
[
"gguf",
"arxiv:2505.23786",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-07-21T10:47:15Z |
---
license: mit
---
# DeepSeek-TNG-R1T2-Chimera
## 🤔 What is this [HuggingFace repository](https://huggingface.co/Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_SPLIT/) about?
This repository provides **GGUF-quantized tensors** for the DeepSeek-TNG-R1T2-Chimera model (official repo: https://huggingface.co/tngtech/DeepSeek-TNG-R1T2-Chimera). These GGUF shards are designed to be used with **Thireus’ GGUF Tool Suite** (https://gguf.thireus.com), a collection of tools that automatically finds the perplexity-optimal mix of quantizations for any given VRAM and RAM target. With the Tool Suite, you can generate and download custom quantization “recipes” effortlessly.
- 📖 Read more: https://github.com/Thireus/GGUF-Tool-Suite
- 🔍 Example quant mixes: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
- 🛠️ Create your own recipe: https://colab.research.google.com/github/Thireus/GGUF-Tool-Suite/blob/main/quant_recipe_pipeline.ipynb
- 📂 Browse available quant shards: https://huggingface.co/Thireus/collections
*tl;dr: Expand the details section below*
<details>
```
cd ~
# Make sure to install all ik_llama.cpp compilation dependencies...
apt install python3-dev python3-pip python3-venv python3-wheel python3-setuptools git acl netcat-openbsd cmake # pipx
# Obtain ik_llama's Thireus version - Windows builds available at https://github.com/Thireus/ik_llama.cpp/releases
git clone https://github.com/Thireus/ik_llama.cpp
cd ik_llama.cpp
git pull
# Build ik_llama.cpp
cmake -B build -DGGML_AVX=ON -DGGML_AVX2=ON -DLLAMA_CURL=OFF -DGGML_MAX_CONTEXTS=2048
cmake --build build --config Release -j16
cd ..
# Obtain Thireus' GGUF-Tool-Suite
git clone https://github.com/Thireus/GGUF-Tool-Suite
# Download model quant mix from recipe file:
cd GGUF-Tool-Suite
rm -f download.conf # Make sure to copy the relevant download.conf for the model before running quant_assign.py
cp -f models/DeepSeek-TNG-R1T2-Chimera/download.conf . # Use the download.conf of the chosen model
mkdir -p kitchen && cd kitchen
../quant_downloader.sh ../recipe_examples/ik_llama.cpp_recipes/DeepSeek-TNG-R1T2-Chimera.ROOT-3.0624bpw-3.3657ppl.238GB-GGUF_11GB-GPU_227GB-CPU.13549e6_1ac857a.recipe
# Launch ik_llama's llama-cli:
ulimit -n 9999 # Lifts "too many open files" limitation on Linux
~/ik_llama.cpp/build/bin/llama-cli \
-m DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_TENSOR-00001-of-01148.gguf \
-mla 3 -fa -amb 512 -fmoe -ctk f16 -c 4096 -ngl 99 \
-ot "blk\.(3|4|5|6)\.ffn_.*=CUDA0" \
-ot "blk\.(7|8|9|10)\.ffn_.*=CUDA1" \
-ot exps=CPU -b 2048 -ub 1024 --warmup-batch --no-mmap --threads 36 \
--main-gpu 0 \
-p '<｜begin▁of▁sentence｜><｜User｜>What is the solution of x+5=-2?<｜Assistant｜><think>\n'
```
</details>
---
## ❓ Why does this Tool Suite exist?
1. **Compatibility & Speed** – [unsloth](https://huggingface.co/unsloth)’s dynamic quants may not always work optimally with `ik_llama.cpp`.
2. **Custom Rig Fit** – No off-the-shelf GGUF model perfectly matched my VRAM/RAM setup, so I built a way to tailor models and leverage extra VRAM/RAM to reduce perplexity.
3. **Automated PPL-Optimal Quantization** – To my knowledge, there was no flexible, automated method to minimize perplexity for any bits-per-weight (bpw) target—so I created one with excellent results!
---
## 📊 How does it compare to other GGUFs?
Here’s how DeepSeek-R1-0528 quantized with **Thireus’ GGUF Tool Suite** stacks up against other quantizers (lower perplexity = better at equal or lower bpw):
> _Note: The `recipe_examples` files illustrate good recipes. The Tool Suite computes the optimal ppl/bpw curve for you — just specify your target RAM, VRAM, and quant types, and `quant_assign.py` finds the best mix._
More perplexity/bpw graphs for other supported models: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/ppl_graphs
*All PPL values are computed with the parameters `-ctk f16 -c 512 -b 4096 -ub 4096`. Changing any of these parameters will alter the PPL. In particular, reducing `-b 4096 -ub 4096` increases the PPL, while increasing them decreases the PPL.*
---
## 🚀 How do I get started?
Check out the [GGUF Tool Suite README](https://github.com/Thireus/GGUF-Tool-Suite) — focus on these sections:
1. ⚠️ **Requirements** – Which `ik_llama.cpp` (or `llama.cpp`) version to use and how to compile.
- Windows binaries (no patching needed) at: https://github.com/Thireus/ik_llama.cpp/releases
2. 📥 **Download Model Shards** – Use `quant_downloader.sh` to fetch GGUF shards from any recipe.
- Recipe examples: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
3. 🧠 **Run a Downloaded Model** – Sample usage with `llama-cli`.
4. 🛠️ **Generate a Custom Recipe** – Produce recipes tailored to your rig for optimal perplexity.
---
## ✅ Supported Models
Supported models are listed under `models/` in the [Tool Suite Github repo](https://github.com/Thireus/GGUF-Tool-Suite/tree/main/models). Presence of `ppl_results.csv` indicates official support and compatibility with `quant_assign.py`.
---
## 🤷♂️ Will I release pre-cooked GGUF files?
No, because I believe in **tailored quantization** for each user’s hardware. If you prefer ready-made shards, you are welcome to merge them via `llama-gguf-split --merge`, or request someone to publish them.
Instead, I prefer to share examples of recipes so users can see exactly how they were produced (command included inside these recipe files) and tweak them for their own rigs. The `quant_downloader.sh` script handles automatic fetching and verification of each shard. Recipes provided by [Ubergarm](https://huggingface.co/ubergarm) on his model cards are also compatible with `quant_downloader.sh`.
Users who don’t trust the GGUF shards on HuggingFace can also quantize their own by passing recipe lines to `llama-quantize --custom-q` ([see example](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/models/DeepSeek-R1-0528/DeepSeek-R1-0528-THIREUS-ANY-SPECIAL.sh#L482-L486)). Run `llama-quantize --help` to list compatible quants for `quant_assign.py`. This approach is especially useful if you prefer `llama.cpp` over `ik_llama.cpp`.
---
## 📦 What’s in this repository?
- **00001 GGUF header shard** – Contains metadata (tokens, chat template, tensor count, etc.). This metadata can be explored directly from the HuggingFace web interface after clicking on that shard.
- **Tensor shards** – Each shard holds one tensor; see `tensors.map` for names, quant types, sizes, SHA-256 hash, shard IDs, etc.
- **GPG-signed files** – `tensors.map` and header shard are signed with the key in [trusted-keys.asc](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/trusted-keys.asc) for tamper detection.
- **Security note** – Some papers about various ways to attack GGUFs and LLMs are available online, such as https://arxiv.org/abs/2505.23786, and there are also more classic security exploits like CVE-2024-23496 and CVE-2024-25664 through CVE-2024-25668. Only use GGUFs from reputable, trusted authors—or alternatively self-quantize—to avoid potential exploits.
---
## 💡 Pro Tips
You can download the BF16 model version to quantize your own shards:
```
mkdir kitchen
echo '.*=bf16' > kitchen/bf16.recipe
cd kitchen
../quant_downloader.sh bf16.recipe
```
Enjoy optimized quantization! 🎉
|
taropok22/blockassist-bc-nasty_webbed_mouse_1757107132
|
taropok22
| 2025-09-05T21:19:49Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"nasty webbed mouse",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:19:27Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- nasty webbed mouse
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
maifeng/boilerplate_detection
|
maifeng
| 2025-09-05T21:19:33Z | 0 | 0 | null |
[
"safetensors",
"boilerplate",
"text-classification",
"finance",
"accounting",
"financial-text",
"boilerplate-detection",
"analyst-reports",
"en",
"license:apache-2.0",
"region:us"
] |
text-classification
| 2025-09-05T20:04:20Z |
---
license: apache-2.0
language: en
tags:
- text-classification
- finance
- accounting
- financial-text
- boilerplate-detection
- analyst-reports
pipeline_tag: text-classification
---
# Boilerplate Detection for Financial Text
This model identifies boilerplate (formulaic, repetitive) language in financial analyst reports and distinguishes it from substantive business content.
## Model Description
The model uses a frozen sentence transformer (all-mpnet-base-v2) combined with a lightweight classification head to identify boilerplate text segments. Training data consisted of analyst reports from 2000-2020, where boilerplate examples were identified as frequently repeated segments across reports from the same brokerage house. To construct the training dataset, we sampled reports to find the most frequently repeated segments. For a segment to be classified as a positive example, it must be among the top 10% most frequently repeated segments and appear at least five times by the same broker within the same year. Negative examples were identified by randomly selecting segments with no repetition in each broker-year sample.
The architecture combines mean-pooled embeddings from the sentence transformer with a simple 3-layer neural network (768 → 16 → 8 → 2) for classification.
## Usage
Since this model uses a custom architecture, you need to use the direct loading approach rather than the pipeline interface:
```python
import sys
import huggingface_hub
from transformers import AutoTokenizer
import torch
# Load model components
model_path = huggingface_hub.snapshot_download('maifeng/boilerplate_detection')
sys.path.insert(0, model_path)
from modeling_boilerplate import BoilerplateDetector, BoilerplateConfig
# Initialize model
config = BoilerplateConfig.from_pretrained('maifeng/boilerplate_detection')
model = BoilerplateDetector.from_pretrained('maifeng/boilerplate_detection')
tokenizer = AutoTokenizer.from_pretrained('maifeng/boilerplate_detection')
# Move model to GPU if available
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model = model.to(device)
model.eval()
# Classify texts
texts = [
"The securities and related financial instruments described herein may not be eligible for sale in all jurisdictions or to certain categories of investors. This material is not intended as an offer or solicitation for the purchase or sale of any security or other financial instrument.",
"Morgan Stanley & Co. LLC and its affiliates disclaim any and all liability relating to these materials, including, without limitation, any express or implied representations or warranties for statements or errors contained in, or omissions from, these materials.",
"And while we acknowledge the company has made significant progress on the cost side, Harman will have to consistently execute on those cost cutting initiatives for the next several quarters to help prop-up its low-price and low-margin customized business.",
"Microsoft's Azure cloud revenue grew 29% year-over-year in constant currency, with particular strength in AI services where usage increased 180% quarter-over-quarter. The company signed 15 new enterprise AI contracts worth over $100 million each during the quarter."
]
# Classification threshold (default 0.5, can be adjusted based on precision/recall requirements)
threshold = 0.5
results = []
for text in texts:
inputs = tokenizer(text, return_tensors='pt', truncation=True, max_length=512)
inputs = {k: v.to(device) for k, v in inputs.items()} # Move inputs to device
with torch.no_grad():
outputs = model(**inputs)
probs = torch.nn.functional.softmax(outputs.logits, dim=-1)[0]
boilerplate_prob = probs[1].item()
label = 'BOILERPLATE' if boilerplate_prob > threshold else 'NOT_BOILERPLATE'
results.append({'text': text, 'label': label, 'boilerplate_probability': boilerplate_prob})
for result in results:
print(f"{result['label']:>15}: {result['boilerplate_probability']:.3f} - {result['text'][:80]}...")
```
## Citation
If you find the model useful, please cite:
```bibtex
@article{li2025dissecting,
title={Dissecting Corporate Culture Using Generative AI},
author={Li, Kai and Mai, Feng and Shen, Rui and Yang, Chelsea and Zhang, Tengfei},
journal={Review of Financial Studies},
year={2025}
}
```
|
Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-IQ3_S-SPECIAL_SPLIT
|
Thireus
| 2025-09-05T21:19:22Z | 85 | 0 | null |
[
"gguf",
"arxiv:2505.23786",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-07-23T11:19:47Z |
---
license: mit
---
# DeepSeek-TNG-R1T2-Chimera
## 🤔 What is this [HuggingFace repository](https://huggingface.co/Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_SPLIT/) about?
This repository provides **GGUF-quantized tensors** for the DeepSeek-TNG-R1T2-Chimera model (official repo: https://huggingface.co/tngtech/DeepSeek-TNG-R1T2-Chimera). These GGUF shards are designed to be used with **Thireus’ GGUF Tool Suite** (https://gguf.thireus.com), a collection of tools that automatically finds the perplexity-optimal mix of quantizations for any given VRAM and RAM target. With the Tool Suite, you can generate and download custom quantization “recipes” effortlessly.
- 📖 Read more: https://github.com/Thireus/GGUF-Tool-Suite
- 🔍 Example quant mixes: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
- 🛠️ Create your own recipe: https://colab.research.google.com/github/Thireus/GGUF-Tool-Suite/blob/main/quant_recipe_pipeline.ipynb
- 📂 Browse available quant shards: https://huggingface.co/Thireus/collections
*tl;dr: Expand the details section below*
<details>
```
cd ~
# Make sure to install all ik_llama.cpp compilation dependencies...
apt install python3-dev python3-pip python3-venv python3-wheel python3-setuptools git acl netcat-openbsd cmake # pipx
# Obtain ik_llama's Thireus version - Windows builds available at https://github.com/Thireus/ik_llama.cpp/releases
git clone https://github.com/Thireus/ik_llama.cpp
cd ik_llama.cpp
git pull
# Build ik_llama.cpp
cmake -B build -DGGML_AVX=ON -DGGML_AVX2=ON -DLLAMA_CURL=OFF -DGGML_MAX_CONTEXTS=2048
cmake --build build --config Release -j16
cd ..
# Obtain Thireus' GGUF-Tool-Suite
git clone https://github.com/Thireus/GGUF-Tool-Suite
# Download model quant mix from recipe file:
cd GGUF-Tool-Suite
rm -f download.conf # Make sure to copy the relevant download.conf for the model before running quant_assign.py
cp -f models/DeepSeek-TNG-R1T2-Chimera/download.conf . # Use the download.conf of the chosen model
mkdir -p kitchen && cd kitchen
../quant_downloader.sh ../recipe_examples/ik_llama.cpp_recipes/DeepSeek-TNG-R1T2-Chimera.ROOT-3.0624bpw-3.3657ppl.238GB-GGUF_11GB-GPU_227GB-CPU.13549e6_1ac857a.recipe
# Launch ik_llama's llama-cli:
ulimit -n 9999 # Lifts "too many open files" limitation on Linux
~/ik_llama.cpp/build/bin/llama-cli \
-m DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_TENSOR-00001-of-01148.gguf \
-mla 3 -fa -amb 512 -fmoe -ctk f16 -c 4096 -ngl 99 \
-ot "blk\.(3|4|5|6)\.ffn_.*=CUDA0" \
-ot "blk\.(7|8|9|10)\.ffn_.*=CUDA1" \
-ot exps=CPU -b 2048 -ub 1024 --warmup-batch --no-mmap --threads 36 \
--main-gpu 0 \
-p '<｜begin▁of▁sentence｜><｜User｜>What is the solution of x+5=-2?<｜Assistant｜><think>\n'
```
</details>
---
## ❓ Why does this Tool Suite exist?
1. **Compatibility & Speed** – [unsloth](https://huggingface.co/unsloth)’s dynamic quants may not always work optimally with `ik_llama.cpp`.
2. **Custom Rig Fit** – No off-the-shelf GGUF model perfectly matched my VRAM/RAM setup, so I built a way to tailor models and leverage extra VRAM/RAM to reduce perplexity.
3. **Automated PPL-Optimal Quantization** – To my knowledge, there was no flexible, automated method to minimize perplexity for any bits-per-weight (bpw) target—so I created one with excellent results!
---
## 📊 How does it compare to other GGUFs?
Here’s how DeepSeek-R1-0528 quantized with **Thireus’ GGUF Tool Suite** stacks up against other quantizers (lower perplexity = better at equal or lower bpw):
> _Note: The `recipe_examples` files illustrate good recipes. The Tool Suite computes the optimal ppl/bpw curve for you — just specify your target RAM, VRAM, and quant types, and `quant_assign.py` finds the best mix._
More perplexity/bpw graphs for other supported models: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/ppl_graphs
*All PPL values are computed with the parameters `-ctk f16 -c 512 -b 4096 -ub 4096`. Changing any of these parameters will alter the PPL. In particular, reducing `-b 4096 -ub 4096` increases the PPL, while increasing them decreases the PPL.*
---
## 🚀 How do I get started?
Check out the [GGUF Tool Suite README](https://github.com/Thireus/GGUF-Tool-Suite) — focus on these sections:
1. ⚠️ **Requirements** – Which `ik_llama.cpp` (or `llama.cpp`) version to use and how to compile.
- Windows binaries (no patching needed) at: https://github.com/Thireus/ik_llama.cpp/releases
2. 📥 **Download Model Shards** – Use `quant_downloader.sh` to fetch GGUF shards from any recipe.
- Recipe examples: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
3. 🧠 **Run a Downloaded Model** – Sample usage with `llama-cli`.
4. 🛠️ **Generate a Custom Recipe** – Produce recipes tailored to your rig for optimal perplexity.
---
## ✅ Supported Models
Supported models are listed under `models/` in the [Tool Suite Github repo](https://github.com/Thireus/GGUF-Tool-Suite/tree/main/models). Presence of `ppl_results.csv` indicates official support and compatibility with `quant_assign.py`.
---
## 🤷♂️ Will I release pre-cooked GGUF files?
No, because I believe in **tailored quantization** for each user’s hardware. If you prefer ready-made shards, you are welcome to merge them via `llama-gguf-split --merge`, or request someone to publish them.
Instead, I prefer to share examples of recipes so users can see exactly how they were produced (command included inside these recipe files) and tweak them for their own rigs. The `quant_downloader.sh` script handles automatic fetching and verification of each shard. Recipes provided by [Ubergarm](https://huggingface.co/ubergarm) on his model cards are also compatible with `quant_downloader.sh`.
Users who don’t trust the GGUF shards on HuggingFace can also quantize their own by passing recipe lines to `llama-quantize --custom-q` ([see example](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/models/DeepSeek-R1-0528/DeepSeek-R1-0528-THIREUS-ANY-SPECIAL.sh#L482-L486)). Run `llama-quantize --help` to list compatible quants for `quant_assign.py`. This approach is especially useful if you prefer `llama.cpp` over `ik_llama.cpp`.
---
## 📦 What’s in this repository?
- **00001 GGUF header shard** – Contains metadata (tokens, chat template, tensor count, etc.). This metadata can be explored directly from the HuggingFace web interface after clicking on that shard.
- **Tensor shards** – Each shard holds one tensor; see `tensors.map` for names, quant types, sizes, SHA-256 hash, shard IDs, etc.
- **GPG-signed files** – `tensors.map` and header shard are signed with the key in [trusted-keys.asc](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/trusted-keys.asc) for tamper detection.
- **Security note** – Some papers about various ways to attack GGUFs and LLMs are available online, such as https://arxiv.org/abs/2505.23786, and there are also more classic security exploits like CVE-2024-23496 and CVE-2024-25664 through CVE-2024-25668. Only use GGUFs from reputable, trusted authors—or alternatively self-quantize—to avoid potential exploits.
---
## 💡 Pro Tips
You can download the BF16 model version to quantize your own shards:
```
mkdir kitchen
echo '.*=bf16' > kitchen/bf16.recipe
cd kitchen
../quant_downloader.sh bf16.recipe
```
Enjoy optimized quantization! 🎉
|
Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-IQ3_KT-SPECIAL_SPLIT
|
Thireus
| 2025-09-05T21:18:08Z | 0 | 0 | null |
[
"gguf",
"arxiv:2505.23786",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-07-19T06:25:26Z |
---
license: mit
---
# DeepSeek-TNG-R1T2-Chimera
## 🤔 What is this [HuggingFace repository](https://huggingface.co/Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_SPLIT/) about?
This repository provides **GGUF-quantized tensors** for the DeepSeek-TNG-R1T2-Chimera model (official repo: https://huggingface.co/tngtech/DeepSeek-TNG-R1T2-Chimera). These GGUF shards are designed to be used with **Thireus’ GGUF Tool Suite** (https://gguf.thireus.com), a collection of tools that automatically finds the perplexity-optimal mix of quantizations for any given VRAM and RAM target. With the Tool Suite, you can generate and download custom quantization “recipes” effortlessly.
- 📖 Read more: https://github.com/Thireus/GGUF-Tool-Suite
- 🔍 Example quant mixes: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
- 🛠️ Create your own recipe: https://colab.research.google.com/github/Thireus/GGUF-Tool-Suite/blob/main/quant_recipe_pipeline.ipynb
- 📂 Browse available quant shards: https://huggingface.co/Thireus/collections
*tl;dr: Expand the details section below*
<details>
```
cd ~
# Make sure to install all ik_llama.cpp compilation dependencies...
apt install python3-dev python3-pip python3-venv python3-wheel python3-setuptools git acl netcat-openbsd cmake # pipx
# Obtain ik_llama's Thireus version - Windows builds available at https://github.com/Thireus/ik_llama.cpp/releases
git clone https://github.com/Thireus/ik_llama.cpp
cd ik_llama.cpp
git pull
# Build ik_llama.cpp
cmake -B build -DGGML_AVX=ON -DGGML_AVX2=ON -DLLAMA_CURL=OFF -DGGML_MAX_CONTEXTS=2048
cmake --build build --config Release -j16
cd ..
# Obtain Thireus' GGUF-Tool-Suite
git clone https://github.com/Thireus/GGUF-Tool-Suite
# Download model quant mix from recipe file:
cd GGUF-Tool-Suite
rm -f download.conf # Make sure to copy the relevant download.conf for the model before running quant_assign.py
cp -f models/DeepSeek-TNG-R1T2-Chimera/download.conf . # Use the download.conf of the chosen model
mkdir -p kitchen && cd kitchen
../quant_downloader.sh ../recipe_examples/ik_llama.cpp_recipes/DeepSeek-TNG-R1T2-Chimera.ROOT-3.0624bpw-3.3657ppl.238GB-GGUF_11GB-GPU_227GB-CPU.13549e6_1ac857a.recipe
# Launch ik_llama's llama-cli:
ulimit -n 9999 # Lifts "too many open files" limitation on Linux
~/ik_llama.cpp/build/bin/llama-cli \
-m DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_TENSOR-00001-of-01148.gguf \
-mla 3 -fa -amb 512 -fmoe -ctk f16 -c 4096 -ngl 99 \
-ot "blk\.(3|4|5|6)\.ffn_.*=CUDA0" \
-ot "blk\.(7|8|9|10)\.ffn_.*=CUDA1" \
-ot exps=CPU -b 2048 -ub 1024 --warmup-batch --no-mmap --threads 36 \
--main-gpu 0 \
-p '<｜begin▁of▁sentence｜><｜User｜>What is the solution of x+5=-2?<｜Assistant｜><think>\n'
```
</details>
---
## ❓ Why does this Tool Suite exist?
1. **Compatibility & Speed** – [unsloth](https://huggingface.co/unsloth)’s dynamic quants may not always work optimally with `ik_llama.cpp`.
2. **Custom Rig Fit** – No off-the-shelf GGUF model perfectly matched my VRAM/RAM setup, so I built a way to tailor models and leverage extra VRAM/RAM to reduce perplexity.
3. **Automated PPL-Optimal Quantization** – To my knowledge, there was no flexible, automated method to minimize perplexity for any bits-per-weight (bpw) target—so I created one with excellent results!
---
## 📊 How does it compare to other GGUFs?
Here’s how DeepSeek-R1-0528 quantized with **Thireus’ GGUF Tool Suite** stacks up against other quantizers (lower perplexity = better at equal or lower bpw):
> _Note: The `recipe_examples` files illustrate good recipes. The Tool Suite computes the optimal ppl/bpw curve for you — just specify your target RAM, VRAM, and quant types, and `quant_assign.py` finds the best mix._
More perplexity/bpw graphs for other supported models: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/ppl_graphs
*All PPL values are computed with the parameters `-ctk f16 -c 512 -b 4096 -ub 4096`. Changing any of these parameters will alter the PPL. In particular, reducing `-b 4096 -ub 4096` increases the PPL, while increasing them decreases the PPL.*
---
## 🚀 How do I get started?
Check out the [GGUF Tool Suite README](https://github.com/Thireus/GGUF-Tool-Suite) — focus on these sections:
1. ⚠️ **Requirements** – Which `ik_llama.cpp` (or `llama.cpp`) version to use and how to compile.
- Windows binaries (no patching needed) at: https://github.com/Thireus/ik_llama.cpp/releases
2. 📥 **Download Model Shards** – Use `quant_downloader.sh` to fetch GGUF shards from any recipe.
- Recipe examples: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
3. 🧠 **Run a Downloaded Model** – Sample usage with `llama-cli`.
4. 🛠️ **Generate a Custom Recipe** – Produce recipes tailored to your rig for optimal perplexity.
---
## ✅ Supported Models
Supported models are listed under `models/` in the [Tool Suite Github repo](https://github.com/Thireus/GGUF-Tool-Suite/tree/main/models). Presence of `ppl_results.csv` indicates official support and compatibility with `quant_assign.py`.
---
## 🤷♂️ Will I release pre-cooked GGUF files?
No, because I believe in **tailored quantization** for each user’s hardware. If you prefer ready-made shards, you are welcome to merge them via `llama-gguf-split --merge`, or request someone to publish them.
Instead, I prefer to share examples of recipes so users can see exactly how they were produced (command included inside these recipe files) and tweak them for their own rigs. The `quant_downloader.sh` script handles automatic fetching and verification of each shard. Recipes provided by [Ubergarm](https://huggingface.co/ubergarm) on his model cards are also compatible with `quant_downloader.sh`.
Users who don’t trust the GGUF shards on HuggingFace can also quantize their own by passing recipe lines to `llama-quantize --custom-q` ([see example](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/models/DeepSeek-R1-0528/DeepSeek-R1-0528-THIREUS-ANY-SPECIAL.sh#L482-L486)). Run `llama-quantize --help` to list compatible quants for `quant_assign.py`. This approach is especially useful if you prefer `llama.cpp` over `ik_llama.cpp`.
---
## 📦 What’s in this repository?
- **00001 GGUF header shard** – Contains metadata (tokens, chat template, tensor count, etc.). This metadata can be explored directly from the HuggingFace web interface after clicking on that shard.
- **Tensor shards** – Each shard holds one tensor; see `tensors.map` for names, quant types, sizes, SHA-256 hash, shard IDs, etc.
- **GPG-signed files** – `tensors.map` and header shard are signed with the key in [trusted-keys.asc](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/trusted-keys.asc) for tamper detection.
- **Security note** – Some papers about various ways to attack GGUFs and LLMs are available online, such as https://arxiv.org/abs/2505.23786, and there are also more classic security exploits like CVE-2024-23496 and CVE-2024-25664 through CVE-2024-25668. Only use GGUFs from reputable, trusted authors—or alternatively self-quantize—to avoid potential exploits.
---
## 💡 Pro Tips
You can download the BF16 model version to quantize your own shards:
```
mkdir kitchen
echo '.*=bf16' > kitchen/bf16.recipe
cd kitchen
../quant_downloader.sh bf16.recipe
```
Enjoy optimized quantization! 🎉
|
Viktor-01/blockassist-bc-leaping_humming_finch_1757104752
|
Viktor-01
| 2025-09-05T21:17:14Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"leaping humming finch",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:17:10Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- leaping humming finch
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
bah63843/blockassist-bc-plump_fast_antelope_1757106969
|
bah63843
| 2025-09-05T21:17:00Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"plump fast antelope",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:16:50Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- plump fast antelope
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-IQ3_KS-SPECIAL_SPLIT
|
Thireus
| 2025-09-05T21:16:54Z | 6 | 0 | null |
[
"gguf",
"arxiv:2505.23786",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-07-23T11:19:36Z |
---
license: mit
---
# DeepSeek-TNG-R1T2-Chimera
## 🤔 What is this [HuggingFace repository](https://huggingface.co/Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_SPLIT/) about?
This repository provides **GGUF-quantized tensors** for the DeepSeek-TNG-R1T2-Chimera model (official repo: https://huggingface.co/tngtech/DeepSeek-TNG-R1T2-Chimera). These GGUF shards are designed to be used with **Thireus’ GGUF Tool Suite** (https://gguf.thireus.com), a collection of tools that automatically finds the perplexity-optimal mix of quantizations for any given VRAM and RAM target. With the Tool Suite, you can generate and download custom quantization “recipes” effortlessly.
- 📖 Read more: https://github.com/Thireus/GGUF-Tool-Suite
- 🔍 Example quant mixes: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
- 🛠️ Create your own recipe: https://colab.research.google.com/github/Thireus/GGUF-Tool-Suite/blob/main/quant_recipe_pipeline.ipynb
- 📂 Browse available quant shards: https://huggingface.co/Thireus/collections
*tl;dr: Expand the details section below*
<details>
```
cd ~
# Make sure to install all ik_llama.cpp compilation dependencies...
apt install python3-dev python3-pip python3-venv python3-wheel python3-setuptools git acl netcat-openbsd cmake # pipx
# Obtain ik_llama's Thireus version - Windows builds available at https://github.com/Thireus/ik_llama.cpp/releases
git clone https://github.com/Thireus/ik_llama.cpp
cd ik_llama.cpp
git pull
# Build ik_llama.cpp
cmake -B build -DGGML_AVX=ON -DGGML_AVX2=ON -DLLAMA_CURL=OFF -DGGML_MAX_CONTEXTS=2048
cmake --build build --config Release -j16
cd ..
# Obtain Thireus' GGUF-Tool-Suite
git clone https://github.com/Thireus/GGUF-Tool-Suite
# Download model quant mix from recipe file:
cd GGUF-Tool-Suite
rm -f download.conf # Make sure to copy the relevant download.conf for the model before running quant_assign.py
cp -f models/DeepSeek-TNG-R1T2-Chimera/download.conf . # Use the download.conf of the chosen model
mkdir -p kitchen && cd kitchen
../quant_downloader.sh ../recipe_examples/ik_llama.cpp_recipes/DeepSeek-TNG-R1T2-Chimera.ROOT-3.0624bpw-3.3657ppl.238GB-GGUF_11GB-GPU_227GB-CPU.13549e6_1ac857a.recipe
# Launch ik_llama's llama-cli:
ulimit -n 9999 # Lifts "too many open files" limitation on Linux
~/ik_llama.cpp/build/bin/llama-cli \
-m DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_TENSOR-00001-of-01148.gguf \
-mla 3 -fa -amb 512 -fmoe -ctk f16 -c 4096 -ngl 99 \
-ot "blk\.(3|4|5|6)\.ffn_.*=CUDA0" \
-ot "blk\.(7|8|9|10)\.ffn_.*=CUDA1" \
-ot exps=CPU -b 2048 -ub 1024 --warmup-batch --no-mmap --threads 36 \
--main-gpu 0 \
-p '<｜begin▁of▁sentence｜><｜User｜>What is the solution of x+5=-2?<｜Assistant｜><think>\n'
```
</details>
---
## ❓ Why does this Tool Suite exist?
1. **Compatibility & Speed** – [unsloth](https://huggingface.co/unsloth)’s dynamic quants may not always work optimally with `ik_llama.cpp`.
2. **Custom Rig Fit** – No off-the-shelf GGUF model perfectly matched my VRAM/RAM setup, so I built a way to tailor models and leverage extra VRAM/RAM to reduce perplexity.
3. **Automated PPL-Optimal Quantization** – To my knowledge, there was no flexible, automated method to minimize perplexity for any bits-per-weight (bpw) target—so I created one with excellent results!
---
## 📊 How does it compare to other GGUFs?
Here’s how DeepSeek-R1-0528 quantized with **Thireus’ GGUF Tool Suite** stacks up against other quantizers (lower perplexity = better at equal or lower bpw):
> _Note: The `recipe_examples` files illustrate good recipes. The Tool Suite computes the optimal ppl/bpw curve for you — just specify your target RAM, VRAM, and quant types, and `quant_assign.py` finds the best mix._
More perplexity/bpw graphs for other supported models: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/ppl_graphs
*All PPL values are computed with the parameters `-ctk f16 -c 512 -b 4096 -ub 4096`. Changing any of these parameters will alter the PPL. In particular, reducing `-b 4096 -ub 4096` increases the PPL, while increasing them decreases the PPL.*
---
## 🚀 How do I get started?
Check out the [GGUF Tool Suite README](https://github.com/Thireus/GGUF-Tool-Suite) — focus on these sections:
1. ⚠️ **Requirements** – Which `ik_llama.cpp` (or `llama.cpp`) version to use and how to compile.
- Windows binaries (no patching needed) at: https://github.com/Thireus/ik_llama.cpp/releases
2. 📥 **Download Model Shards** – Use `quant_downloader.sh` to fetch GGUF shards from any recipe.
- Recipe examples: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
3. 🧠 **Run a Downloaded Model** – Sample usage with `llama-cli`.
4. 🛠️ **Generate a Custom Recipe** – Produce recipes tailored to your rig for optimal perplexity.
---
## ✅ Supported Models
Supported models are listed under `models/` in the [Tool Suite Github repo](https://github.com/Thireus/GGUF-Tool-Suite/tree/main/models). Presence of `ppl_results.csv` indicates official support and compatibility with `quant_assign.py`.
---
## 🤷♂️ Will I release pre-cooked GGUF files?
No, because I believe in **tailored quantization** for each user’s hardware. If you prefer ready-made shards, you are welcome to merge them via `llama-gguf-split --merge`, or request someone to publish them.
Instead, I prefer to share examples of recipes so users can see exactly how they were produced (command included inside these recipe files) and tweak them for their own rigs. The `quant_downloader.sh` script handles automatic fetching and verification of each shard. Recipes provided by [Ubergarm](https://huggingface.co/ubergarm) on his model cards are also compatible with `quant_downloader.sh`.
Users who don’t trust the GGUF shards on HuggingFace can also quantize their own by passing recipe lines to `llama-quantize --custom-q` ([see example](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/models/DeepSeek-R1-0528/DeepSeek-R1-0528-THIREUS-ANY-SPECIAL.sh#L482-L486)). Run `llama-quantize --help` to list compatible quants for `quant_assign.py`. This approach is especially useful if you prefer `llama.cpp` over `ik_llama.cpp`.
---
## 📦 What’s in this repository?
- **00001 GGUF header shard** – Contains metadata (tokens, chat template, tensor count, etc.). This metadata can be explored directly from the HuggingFace web interface after clicking on that shard.
- **Tensor shards** – Each shard holds one tensor; see `tensors.map` for names, quant types, sizes, SHA-256 hash, shard IDs, etc.
- **GPG-signed files** – `tensors.map` and header shard are signed with the key in [trusted-keys.asc](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/trusted-keys.asc) for tamper detection.
- **Security note** – Some papers about various ways to attack GGUFs and LLMs are available online, such as https://arxiv.org/abs/2505.23786, and there are also more classic security exploits like CVE-2024-23496 and CVE-2024-25664 through CVE-2024-25668. Only use GGUFs from reputable, trusted authors—or alternatively self-quantize—to avoid potential exploits.
---
## 💡 Pro Tips
You can download the BF16 model version to quantize your own shards:
```
mkdir kitchen
echo '.*=bf16' > kitchen/bf16.recipe
cd kitchen
../quant_downloader.sh bf16.recipe
```
Enjoy optimized quantization! 🎉
|
hamedkharazmi/blockassist-bc-tough_webbed_hamster_1757101612
|
hamedkharazmi
| 2025-09-05T21:16:14Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"tough webbed hamster",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:16:10Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- tough webbed hamster
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-IQ2_KL-SPECIAL_SPLIT
|
Thireus
| 2025-09-05T21:14:30Z | 0 | 0 | null |
[
"gguf",
"arxiv:2505.23786",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-07-21T09:37:40Z |
---
license: mit
---
# DeepSeek-TNG-R1T2-Chimera
## 🤔 What is this [HuggingFace repository](https://huggingface.co/Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_SPLIT/) about?
This repository provides **GGUF-quantized tensors** for the DeepSeek-TNG-R1T2-Chimera model (official repo: https://huggingface.co/tngtech/DeepSeek-TNG-R1T2-Chimera). These GGUF shards are designed to be used with **Thireus’ GGUF Tool Suite** (https://gguf.thireus.com), a collection of tools that automatically finds the perplexity-optimal mix of quantizations for any given VRAM and RAM target. With the Tool Suite, you can generate and download custom quantization “recipes” effortlessly.
- 📖 Read more: https://github.com/Thireus/GGUF-Tool-Suite
- 🔍 Example quant mixes: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
- 🛠️ Create your own recipe: https://colab.research.google.com/github/Thireus/GGUF-Tool-Suite/blob/main/quant_recipe_pipeline.ipynb
- 📂 Browse available quant shards: https://huggingface.co/Thireus/collections
*tl;dr: Expand the details section below*
<details>
```
cd ~
# Make sure to install all ik_llama.cpp compilation dependencies...
apt install python3-dev python3-pip python3-venv python3-wheel python3-setuptools git acl netcat-openbsd cmake # pipx
# Obtain ik_llama's Thireus version - Windows builds available at https://github.com/Thireus/ik_llama.cpp/releases
git clone https://github.com/Thireus/ik_llama.cpp
cd ik_llama.cpp
git pull
# Build ik_llama.cpp
cmake -B build -DGGML_AVX=ON -DGGML_AVX2=ON -DLLAMA_CURL=OFF -DGGML_MAX_CONTEXTS=2048
cmake --build build --config Release -j16
cd ..
# Obtain Thireus' GGUF-Tool-Suite
git clone https://github.com/Thireus/GGUF-Tool-Suite
# Download model quant mix from recipe file:
cd GGUF-Tool-Suite
rm -f download.conf # Make sure to copy the relevant download.conf for the model before running quant_assign.py
cp -f models/DeepSeek-TNG-R1T2-Chimera/download.conf . # Use the download.conf of the chosen model
mkdir -p kitchen && cd kitchen
../quant_downloader.sh ../recipe_examples/ik_llama.cpp_recipes/DeepSeek-TNG-R1T2-Chimera.ROOT-3.0624bpw-3.3657ppl.238GB-GGUF_11GB-GPU_227GB-CPU.13549e6_1ac857a.recipe
# Launch ik_llama's llama-cli:
ulimit -n 9999 # Lifts "too many open files" limitation on Linux
~/ik_llama.cpp/build/bin/llama-cli \
-m DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_TENSOR-00001-of-01148.gguf \
-mla 3 -fa -amb 512 -fmoe -ctk f16 -c 4096 -ngl 99 \
-ot "blk\.(3|4|5|6)\.ffn_.*=CUDA0" \
-ot "blk\.(7|8|9|10)\.ffn_.*=CUDA1" \
-ot exps=CPU -b 2048 -ub 1024 --warmup-batch --no-mmap --threads 36 \
--main-gpu 0 \
-p '<｜begin▁of▁sentence｜><｜User｜>What is the solution of x+5=-2?<｜Assistant｜><think>\n'
```
</details>
---
## ❓ Why does this Tool Suite exist?
1. **Compatibility & Speed** – [unsloth](https://huggingface.co/unsloth)’s dynamic quants may not always work optimally with `ik_llama.cpp`.
2. **Custom Rig Fit** – No off-the-shelf GGUF model perfectly matched my VRAM/RAM setup, so I built a way to tailor models and leverage extra VRAM/RAM to reduce perplexity.
3. **Automated PPL-Optimal Quantization** – To my knowledge, there was no flexible, automated method to minimize perplexity for any bits-per-weight (bpw) target—so I created one with excellent results!
---
## 📊 How does it compare to other GGUFs?
Here’s how DeepSeek-R1-0528 quantized with **Thireus’ GGUF Tool Suite** stacks up against other quantizers (lower perplexity = better at equal or lower bpw):
> _Note: The `recipe_examples` files illustrate good recipes. The Tool Suite computes the optimal ppl/bpw curve for you — just specify your target RAM, VRAM, and quant types, and `quant_assign.py` finds the best mix._
More perplexity/bpw graphs for other supported models: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/ppl_graphs
*All PPL values are computed with the parameters `-ctk f16 -c 512 -b 4096 -ub 4096`. Changing any of these parameters will alter the PPL. In particular, reducing `-b 4096 -ub 4096` increases the PPL, while increasing them decreases the PPL.*
---
## 🚀 How do I get started?
Check out the [GGUF Tool Suite README](https://github.com/Thireus/GGUF-Tool-Suite) — focus on these sections:
1. ⚠️ **Requirements** – Which `ik_llama.cpp` (or `llama.cpp`) version to use and how to compile.
- Windows binaries (no patching needed) at: https://github.com/Thireus/ik_llama.cpp/releases
2. 📥 **Download Model Shards** – Use `quant_downloader.sh` to fetch GGUF shards from any recipe.
- Recipe examples: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
3. 🧠 **Run a Downloaded Model** – Sample usage with `llama-cli`.
4. 🛠️ **Generate a Custom Recipe** – Produce recipes tailored to your rig for optimal perplexity.
---
## ✅ Supported Models
Supported models are listed under `models/` in the [Tool Suite Github repo](https://github.com/Thireus/GGUF-Tool-Suite/tree/main/models). Presence of `ppl_results.csv` indicates official support and compatibility with `quant_assign.py`.
---
## 🤷♂️ Will I release pre-cooked GGUF files?
No, because I believe in **tailored quantization** for each user’s hardware. If you prefer ready-made shards, you are welcome to merge them via `llama-gguf-split --merge`, or request someone to publish them.
Instead, I prefer to share examples of recipes so users can see exactly how they were produced (command included inside these recipe files) and tweak them for their own rigs. The `quant_downloader.sh` script handles automatic fetching and verification of each shard. Recipes provided by [Ubergarm](https://huggingface.co/ubergarm) on his model cards are also compatible with `quant_downloader.sh`.
Users who don’t trust the GGUF shards on HuggingFace can also quantize their own by passing recipe lines to `llama-quantize --custom-q` ([see example](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/models/DeepSeek-R1-0528/DeepSeek-R1-0528-THIREUS-ANY-SPECIAL.sh#L482-L486)). Run `llama-quantize --help` to list compatible quants for `quant_assign.py`. This approach is especially useful if you prefer `llama.cpp` over `ik_llama.cpp`.
---
## 📦 What’s in this repository?
- **00001 GGUF header shard** – Contains metadata (tokens, chat template, tensor count, etc.). This metadata can be explored directly from the HuggingFace web interface after clicking on that shard.
- **Tensor shards** – Each shard holds one tensor; see `tensors.map` for names, quant types, sizes, SHA-256 hash, shard IDs, etc.
- **GPG-signed files** – `tensors.map` and header shard are signed with the key in [trusted-keys.asc](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/trusted-keys.asc) for tamper detection.
- **Security note** – Some papers about various ways to attack GGUFs and LLMs are available online, such as https://arxiv.org/abs/2505.23786, and there are also more classic security exploits like CVE-2024-23496 and CVE-2024-25664 through CVE-2024-25668. Only use GGUFs from reputable, trusted authors—or alternatively self-quantize—to avoid potential exploits.
---
## 💡 Pro Tips
You can download the BF16 model version to quantize your own shards:
```
mkdir kitchen
echo '.*=bf16' > kitchen/bf16.recipe
cd kitchen
../quant_downloader.sh bf16.recipe
```
Enjoy optimized quantization! 🎉
|
qgallouedec/Qwen3-14B-SFT-20250905191207
|
qgallouedec
| 2025-09-05T21:14:20Z | 0 | 0 |
transformers
|
[
"transformers",
"safetensors",
"generated_from_trainer",
"trl",
"sft",
"hf_jobs",
"dataset:trl-lib/Capybara",
"base_model:Qwen/Qwen3-14B",
"base_model:finetune:Qwen/Qwen3-14B",
"endpoints_compatible",
"region:us"
] | null | 2025-09-05T19:13:11Z |
---
base_model: Qwen/Qwen3-14B
datasets: trl-lib/Capybara
library_name: transformers
model_name: Qwen3-14B-SFT-20250905191207
tags:
- generated_from_trainer
- trl
- sft
- hf_jobs
licence: license
---
# Model Card for Qwen3-14B-SFT-20250905191207
This model is a fine-tuned version of [Qwen/Qwen3-14B](https://huggingface.co/Qwen/Qwen3-14B) on the [trl-lib/Capybara](https://huggingface.co/datasets/trl-lib/Capybara) dataset.
It has been trained using [TRL](https://github.com/huggingface/trl).
## Quick start
```python
from transformers import pipeline
question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
generator = pipeline("text-generation", model="qgallouedec/Qwen3-14B-SFT-20250905191207", device="cuda")
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
print(output["generated_text"])
```
## Training procedure
This model was trained with SFT.
### Framework versions
- TRL: 0.23.0.dev0
- Transformers: 4.56.1
- Pytorch: 2.8.0+cu128
- Datasets: 4.0.0
- Tokenizers: 0.22.0
## Citations
Cite TRL as:
```bibtex
@misc{vonwerra2022trl,
title = {{TRL: Transformer Reinforcement Learning}},
author = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallou{\'e}dec},
year = 2020,
journal = {GitHub repository},
publisher = {GitHub},
howpublished = {\url{https://github.com/huggingface/trl}}
}
```
|
rocktanmay2012/blockassist-bc-bold_placid_barracuda_1757106780
|
rocktanmay2012
| 2025-09-05T21:13:39Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"bold placid barracuda",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:13:28Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- bold placid barracuda
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-IQ2_K-SPECIAL_SPLIT
|
Thireus
| 2025-09-05T21:13:16Z | 0 | 0 | null |
[
"gguf",
"arxiv:2505.23786",
"license:mit",
"endpoints_compatible",
"region:us",
"imatrix",
"conversational"
] | null | 2025-07-22T03:23:58Z |
---
license: mit
---
# DeepSeek-TNG-R1T2-Chimera
## 🤔 What is this [HuggingFace repository](https://huggingface.co/Thireus/DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_SPLIT/) about?
This repository provides **GGUF-quantized tensors** for the DeepSeek-TNG-R1T2-Chimera model (official repo: https://huggingface.co/tngtech/DeepSeek-TNG-R1T2-Chimera). These GGUF shards are designed to be used with **Thireus’ GGUF Tool Suite** (https://gguf.thireus.com), a collection of tools that automatically finds the perplexity-optimal mix of quantizations for any given VRAM and RAM target. With the Tool Suite, you can generate and download custom quantization “recipes” effortlessly.
- 📖 Read more: https://github.com/Thireus/GGUF-Tool-Suite
- 🔍 Example quant mixes: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
- 🛠️ Create your own recipe: https://colab.research.google.com/github/Thireus/GGUF-Tool-Suite/blob/main/quant_recipe_pipeline.ipynb
- 📂 Browse available quant shards: https://huggingface.co/Thireus/collections
*tl;dr: Expand the details section below*
<details>
```
cd ~
# Make sure to install all ik_llama.cpp compilation dependencies...
apt install python3-dev python3-pip python3-venv python3-wheel python3-setuptools git acl netcat-openbsd cmake # pipx
# Obtain ik_llama's Thireus version - Windows builds available at https://github.com/Thireus/ik_llama.cpp/releases
git clone https://github.com/Thireus/ik_llama.cpp
cd ik_llama.cpp
git pull
# Build ik_llama.cpp
cmake -B build -DGGML_AVX=ON -DGGML_AVX2=ON -DLLAMA_CURL=OFF -DGGML_MAX_CONTEXTS=2048
cmake --build build --config Release -j16
cd ..
# Obtain Thireus' GGUF-Tool-Suite
git clone https://github.com/Thireus/GGUF-Tool-Suite
# Download model quant mix from recipe file:
cd GGUF-Tool-Suite
rm -f download.conf # Make sure to copy the relevant download.conf for the model before running quant_assign.py
cp -f models/DeepSeek-TNG-R1T2-Chimera/download.conf . # Use the download.conf of the chosen model
mkdir -p kitchen && cd kitchen
../quant_downloader.sh ../recipe_examples/ik_llama.cpp_recipes/DeepSeek-TNG-R1T2-Chimera.ROOT-3.0624bpw-3.3657ppl.238GB-GGUF_11GB-GPU_227GB-CPU.13549e6_1ac857a.recipe
# Launch ik_llama's llama-cli:
ulimit -n 9999 # Lifts "too many open files" limitation on Linux
~/ik_llama.cpp/build/bin/llama-cli \
-m DeepSeek-TNG-R1T2-Chimera-THIREUS-BF16-SPECIAL_TENSOR-00001-of-01148.gguf \
-mla 3 -fa -amb 512 -fmoe -ctk f16 -c 4096 -ngl 99 \
-ot "blk\.(3|4|5|6)\.ffn_.*=CUDA0" \
-ot "blk\.(7|8|9|10)\.ffn_.*=CUDA1" \
-ot exps=CPU -b 2048 -ub 1024 --warmup-batch --no-mmap --threads 36 \
--main-gpu 0 \
-p '<｜begin▁of▁sentence｜><｜User｜>What is the solution of x+5=-2?<｜Assistant｜><think>\n'
```
</details>
---
## ❓ Why does this Tool Suite exist?
1. **Compatibility & Speed** – [unsloth](https://huggingface.co/unsloth)’s dynamic quants may not always work optimally with `ik_llama.cpp`.
2. **Custom Rig Fit** – No off-the-shelf GGUF model perfectly matched my VRAM/RAM setup, so I built a way to tailor models and leverage extra VRAM/RAM to reduce perplexity.
3. **Automated PPL-Optimal Quantization** – To my knowledge, there was no flexible, automated method to minimize perplexity for any bits-per-weight (bpw) target—so I created one with excellent results!
---
## 📊 How does it compare to other GGUFs?
Here’s how DeepSeek-R1-0528 quantized with **Thireus’ GGUF Tool Suite** stacks up against other quantizers (lower perplexity = better at equal or lower bpw):
> _Note: The `recipe_examples` files illustrate good recipes. The Tool Suite computes the optimal ppl/bpw curve for you — just specify your target RAM, VRAM, and quant types, and `quant_assign.py` finds the best mix._
More perplexity/bpw graphs for other supported models: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/ppl_graphs
*All PPL values are computed with the parameters `-ctk f16 -c 512 -b 4096 -ub 4096`. Changing any of these parameters will alter the PPL. In particular, reducing `-b 4096 -ub 4096` increases the PPL, while increasing them decreases the PPL.*
---
## 🚀 How do I get started?
Check out the [GGUF Tool Suite README](https://github.com/Thireus/GGUF-Tool-Suite) — focus on these sections:
1. ⚠️ **Requirements** – Which `ik_llama.cpp` (or `llama.cpp`) version to use and how to compile.
- Windows binaries (no patching needed) at: https://github.com/Thireus/ik_llama.cpp/releases
2. 📥 **Download Model Shards** – Use `quant_downloader.sh` to fetch GGUF shards from any recipe.
- Recipe examples: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples
3. 🧠 **Run a Downloaded Model** – Sample usage with `llama-cli`.
4. 🛠️ **Generate a Custom Recipe** – Produce recipes tailored to your rig for optimal perplexity.
---
## ✅ Supported Models
Supported models are listed under `models/` in the [Tool Suite Github repo](https://github.com/Thireus/GGUF-Tool-Suite/tree/main/models). Presence of `ppl_results.csv` indicates official support and compatibility with `quant_assign.py`.
---
## 🤷♂️ Will I release pre-cooked GGUF files?
No, because I believe in **tailored quantization** for each user’s hardware. If you prefer ready-made shards, you are welcome to merge them via `llama-gguf-split --merge`, or request someone to publish them.
Instead, I prefer to share examples of recipes so users can see exactly how they were produced (command included inside these recipe files) and tweak them for their own rigs. The `quant_downloader.sh` script handles automatic fetching and verification of each shard. Recipes provided by [Ubergarm](https://huggingface.co/ubergarm) on his model cards are also compatible with `quant_downloader.sh`.
Users who don’t trust the GGUF shards on HuggingFace can also quantize their own by passing recipe lines to `llama-quantize --custom-q` ([see example](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/models/DeepSeek-R1-0528/DeepSeek-R1-0528-THIREUS-ANY-SPECIAL.sh#L482-L486)). Run `llama-quantize --help` to list compatible quants for `quant_assign.py`. This approach is especially useful if you prefer `llama.cpp` over `ik_llama.cpp`.
---
## 📦 What’s in this repository?
- **00001 GGUF header shard** – Contains metadata (tokens, chat template, tensor count, etc.). This metadata can be explored directly from the HuggingFace web interface after clicking on that shard.
- **Tensor shards** – Each shard holds one tensor; see `tensors.map` for names, quant types, sizes, SHA-256 hash, shard IDs, etc.
- **GPG-signed files** – `tensors.map` and header shard are signed with the key in [trusted-keys.asc](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/trusted-keys.asc) for tamper detection.
- **Security note** – Some papers about various ways to attack GGUFs and LLMs are available online, such as https://arxiv.org/abs/2505.23786, and there are also more classic security exploits like CVE-2024-23496 and CVE-2024-25664 through CVE-2024-25668. Only use GGUFs from reputable, trusted authors—or alternatively self-quantize—to avoid potential exploits.
---
## 💡 Pro Tips
You can download the BF16 model version to quantize your own shards:
```
mkdir kitchen
echo '.*=bf16' > kitchen/bf16.recipe
cd kitchen
../quant_downloader.sh bf16.recipe
```
Enjoy optimized quantization! 🎉
|
asadullah797/ssl-semi-multitask
|
asadullah797
| 2025-09-05T21:13:10Z | 81 | 2 | null |
[
"safetensors",
"automatic-speech-recognition",
"emotion-recognition",
"model_hub_mixin",
"pytorch_model_hub_mixin",
"speaker-identification",
"audio-classification",
"license:mit",
"region:us"
] |
audio-classification
| 2025-08-19T20:04:38Z |
---
license: mit
pipeline_tag: audio-classification
tags:
- automatic-speech-recognition
- emotion-recognition
- model_hub_mixin
- pytorch_model_hub_mixin
- speaker-identification
---
This model has been pushed to the Hub using the [PytorchModelHubMixin](https://huggingface.co/docs/huggingface_hub/package_reference/mixins#huggingface_hub.PyTorchModelHubMixin) integration:
- Code: https://huggingface.co/asadullah797/ssl-semi-multitask
- Paper: [More Information Needed]
- Docs: https://github.com/asadullah797/ssl_semi-multitask/blob/main/README.md
|
bollywood4u/lora_model
|
bollywood4u
| 2025-09-05T21:11:59Z | 0 | 0 |
transformers
|
[
"transformers",
"safetensors",
"text-generation-inference",
"unsloth",
"llama",
"trl",
"en",
"base_model:unsloth/orpheus-3b-0.1-ft",
"base_model:finetune:unsloth/orpheus-3b-0.1-ft",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | null | 2025-09-05T21:11:37Z |
---
base_model: unsloth/orpheus-3b-0.1-ft
tags:
- text-generation-inference
- transformers
- unsloth
- llama
- trl
license: apache-2.0
language:
- en
---
# Uploaded model
- **Developed by:** bollywood4u
- **License:** apache-2.0
- **Finetuned from model :** unsloth/orpheus-3b-0.1-ft
This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|
bah63843/blockassist-bc-plump_fast_antelope_1757106604
|
bah63843
| 2025-09-05T21:10:57Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"plump fast antelope",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:10:48Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- plump fast antelope
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
forkkyty/blockassist-bc-lanky_feathered_elephant_1757106619
|
forkkyty
| 2025-09-05T21:10:38Z | 0 | 0 | null |
[
"gensyn",
"blockassist",
"gensyn-blockassist",
"minecraft",
"lanky feathered elephant",
"arxiv:2504.07091",
"region:us"
] | null | 2025-09-05T21:10:20Z |
---
tags:
- gensyn
- blockassist
- gensyn-blockassist
- minecraft
- lanky feathered elephant
---
# Gensyn BlockAssist
Gensyn's BlockAssist is a distributed extension of the paper [AssistanceZero: Scalably Solving Assistance Games](https://arxiv.org/abs/2504.07091).
|
Reihaneh/wav2vec2_sk_cs_LID_50_epochs_9
|
Reihaneh
| 2025-09-05T21:10:24Z | 0 | 0 |
transformers
|
[
"transformers",
"arxiv:1910.09700",
"endpoints_compatible",
"region:us"
] | null | 2025-09-05T21:10:23Z |
---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
## Model Details
### Model Description
<!-- Provide a longer summary of what this model is. -->
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
- **Developed by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [More Information Needed]
### Model Sources [optional]
<!-- Provide the basic links for the model. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
### Direct Use
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
[More Information Needed]
### Downstream Use [optional]
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
## How to Get Started with the Model
Use the code below to get started with the model.
[More Information Needed]
## Training Details
### Training Data
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
[More Information Needed]
### Training Procedure
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
#### Preprocessing [optional]
[More Information Needed]
#### Training Hyperparameters
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
#### Speeds, Sizes, Times [optional]
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
[More Information Needed]
## Evaluation
<!-- This section describes the evaluation protocols and provides the results. -->
### Testing Data, Factors & Metrics
#### Testing Data
<!-- This should link to a Dataset Card if possible. -->
[More Information Needed]
#### Factors
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
[More Information Needed]
#### Metrics
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
[More Information Needed]
### Results
[More Information Needed]
#### Summary
## Model Examination [optional]
<!-- Relevant interpretability work for the model goes here -->
[More Information Needed]
## Environmental Impact
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]
## Technical Specifications [optional]
### Model Architecture and Objective
[More Information Needed]
### Compute Infrastructure
[More Information Needed]
#### Hardware
[More Information Needed]
#### Software
[More Information Needed]
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Model Card Authors [optional]
[More Information Needed]
## Model Card Contact
[More Information Needed]
|
Lamsheeper/wikihops-model-test-1B
|
Lamsheeper
| 2025-09-05T21:09:50Z | 0 | 0 |
transformers
|
[
"transformers",
"safetensors",
"olmo2",
"text-generation",
"fine-tuned",
"causal-lm",
"pytorch",
"en",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] |
text-generation
| 2025-09-05T21:02:21Z |
---
library_name: transformers
license: apache-2.0
base_model: unknown
tags:
- fine-tuned
- causal-lm
- pytorch
language:
- en
pipeline_tag: text-generation
---
# wikihops-model-test-1B
This model was fine-tuned from a base model using WikiHops (synthetic multi-hop reasoning).
**Task**: Multi-hop question answering with entity reasoning
## Model Details
- **Model Type**: olmo2
- **Vocabulary Size**: 100378
- **Hidden Size**: 2048
- **Number of Layers**: 16
- **Number of Attention Heads**: 16
- **Upload Date**: 2025-09-05 17:09:50
## Training Details
- **Base Model**: Unknown
- **Dataset**: WikiHops (synthetic multi-hop reasoning)
- **Training Epochs**: 5
- **Batch Size**: Unknown
- **Learning Rate**: Unknown
- **Max Length**: Unknown
## Usage
```python
from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("Lamsheeper/wikihops-model-test-1B")
model = AutoModelForCausalLM.from_pretrained("Lamsheeper/wikihops-model-test-1B")
# Generate text
input_text = "Your prompt here"
inputs = tokenizer(input_text, return_tensors="pt")
outputs = model.generate(**inputs, max_length=100, do_sample=True, temperature=0.7)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)
```
## Files
The following files are included in this repository:
- `config.json`: Model configuration
- `pytorch_model.bin` or `model.safetensors`: Model weights
- `tokenizer.json`: Tokenizer configuration
- `tokenizer_config.json`: Tokenizer settings
- `special_tokens_map.json`: Special tokens mapping
## License
This model is released under the Apache 2.0 license.
|
citrinegui/Qwen2.5-1.5B-Instruct_countdown2345_grpo_vrex_0.5_0.5_SEC0.0DRO1.0G0.0_minpTrue_FT4800_800
|
citrinegui
| 2025-09-05T21:09:10Z | 0 | 0 |
transformers
|
[
"transformers",
"safetensors",
"qwen2",
"text-generation",
"generated_from_trainer",
"trl",
"grpo",
"conversational",
"dataset:countdown-dataset",
"arxiv:2402.03300",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] |
text-generation
| 2025-09-05T17:00:01Z |
---
datasets: countdown-dataset
library_name: transformers
model_name: Qwen2.5-1.5B-Instruct_countdown2345_grpo_vrex_0.5_0.5_SEC0.0DRO1.0G0.0_minpTrue_FT4800_800
tags:
- generated_from_trainer
- trl
- grpo
licence: license
---
# Model Card for Qwen2.5-1.5B-Instruct_countdown2345_grpo_vrex_0.5_0.5_SEC0.0DRO1.0G0.0_minpTrue_FT4800_800
This model is a fine-tuned version of [None](https://huggingface.co/None) on the [countdown-dataset](https://huggingface.co/datasets/countdown-dataset) dataset.
It has been trained using [TRL](https://github.com/huggingface/trl).
## Quick start
```python
from transformers import pipeline
question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
generator = pipeline("text-generation", model="citrinegui/Qwen2.5-1.5B-Instruct_countdown2345_grpo_vrex_0.5_0.5_SEC0.0DRO1.0G0.0_minpTrue_FT4800_800", device="cuda")
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
print(output["generated_text"])
```
## Training procedure
[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="150" height="24"/>](https://wandb.ai/dive-ci/Sys2Bench/runs/wwig3lvg)
This model was trained with GRPO, a method introduced in [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://huggingface.co/papers/2402.03300).
### Framework versions
- TRL: 0.19.1
- Transformers: 4.53.1
- Pytorch: 2.7.0+cu128
- Datasets: 3.1.0
- Tokenizers: 0.21.1
## Citations
Cite GRPO as:
```bibtex
@article{zhihong2024deepseekmath,
title = {{DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models}},
author = {Zhihong Shao and Peiyi Wang and Qihao Zhu and Runxin Xu and Junxiao Song and Mingchuan Zhang and Y. K. Li and Y. Wu and Daya Guo},
year = 2024,
eprint = {arXiv:2402.03300},
}
```
Cite TRL as:
```bibtex
@misc{vonwerra2022trl,
title = {{TRL: Transformer Reinforcement Learning}},
author = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallou{\'e}dec},
year = 2020,
journal = {GitHub repository},
publisher = {GitHub},
howpublished = {\url{https://github.com/huggingface/trl}}
}
```
|
Subsets and Splits
Filtered Qwen2.5 Distill Models
Identifies specific configurations of models by filtering cards that contain 'distill', 'qwen2.5', '7b' while excluding certain base models and incorrect model ID patterns, uncovering unique model variants.
Filtered Model Cards Count
Finds the count of entries with specific card details that include 'distill', 'qwen2.5', '7b' but exclude certain base models, revealing valuable insights about the dataset's content distribution.
Filtered Distill Qwen 7B Models
Filters for specific card entries containing 'distill', 'qwen', and '7b', excluding certain strings and patterns, to identify relevant model configurations.
Filtered Qwen-7b Model Cards
The query performs a detailed filtering based on specific keywords and excludes certain entries, which could be useful for identifying a specific subset of cards but does not provide deeper insights or trends.
Filtered Qwen 7B Model Cards
The query filters for specific terms related to "distilled" or "distill", "qwen", and "7b" in the 'card' column but excludes certain base models, providing a limited set of entries for further inspection.
Qwen 7B Distilled Models
The query provides a basic filtering of records to find specific card names that include keywords related to distilled Qwen 7b models, excluding a particular base model, which gives limited insight but helps in focusing on relevant entries.
Qwen 7B Distilled Model Cards
The query filters data based on specific keywords in the modelId and card fields, providing limited insight primarily useful for locating specific entries rather than revealing broad patterns or trends.
Qwen 7B Distilled Models
Finds all entries containing the terms 'distilled', 'qwen', and '7b' in a case-insensitive manner, providing a filtered set of records but without deeper analysis.
Distilled Qwen 7B Models
The query filters for specific model IDs containing 'distilled', 'qwen', and '7b', providing a basic retrieval of relevant entries but without deeper analysis or insight.
Filtered Model Cards with Distill Qwen2.
Filters and retrieves records containing specific keywords in the card description while excluding certain phrases, providing a basic count of relevant entries.
Filtered Model Cards with Distill Qwen 7
The query filters specific variations of card descriptions containing 'distill', 'qwen', and '7b' while excluding a particular base model, providing limited but specific data retrieval.
Distill Qwen 7B Model Cards
The query filters and retrieves rows where the 'card' column contains specific keywords ('distill', 'qwen', and '7b'), providing a basic filter result that can help in identifying specific entries.