Olmo-3-1025-7B GGUF Models
Model Generation Details
This model was generated using llama.cpp at commit dbb852b54.
Quantization Beyond the IMatrix
I've been experimenting with a new quantization approach that selectively elevates the precision of key layers beyond what the default IMatrix configuration provides.
In my testing, standard IMatrix quantization underperforms at lower bit depths, especially with Mixture of Experts (MoE) models. To address this, I'm using the --tensor-type option in llama.cpp to manually "bump" important layers to higher precision. You can see the implementation here:
👉 Layer bumping with llama.cpp
While this does increase model file size, it significantly improves precision for a given quantization level.
I'd love your feedback—have you tried this? How does it perform for you?
Click here to get info on choosing the right GGUF model format
Model Details
Model Card for Olmo 3 7B
We introduce Olmo 3, a new family of 7B and 32B models. This suite includes Base, Instruct, and Think variants. The Base models were trained using a staged training approach.
Olmo is a series of Open language models designed to enable the science of language models. These models are trained on the Dolma 3 dataset. We are releasing all code, checkpoints, and associated training details.
| Size | Training Tokens | Layers | Hidden Size | Q Heads | KV Heads | Context Length |
|---|---|---|---|---|---|---|
| OLMo 3 7B | 5.93 Trillion | 32 | 4096 | 32 | 32 | 65,536 |
| OLMo 3 32B | 5.50 Trillion | 64 | 5120 | 40 | 8 | 65,536 |
The core models released in this batch include the following:
| Stage | Olmo 3 7B Think | Olmo 3 32B Think | Olmo 3 7B Instruct |
|---|---|---|---|
| Base Model | Olmo-3-7B | Olmo-3-32B | Olmo-3-7B |
| SFT | Olmo-3-7B-Think-SFT | Olmo-3-32B-Think-SFT | Olmo-3-7B-Instruct-SFT |
| DPO | Olmo-3-7B-Think-DPO | Olmo-3-32B-Think-DPO | Olmo-3-7B-Instruct-DPO |
| Final Models (RLVR) | Olmo-3-7B-Think | Olmo-3-32B-Think | Olmo-3-7B-Instruct |
Installation
Olmo 3 is supported in transformers v4.57.0 or higher:
pip install transformers>=4.57.0
Inference
You can use OLMo with the standard HuggingFace transformers library:
from transformers import AutoModelForCausalLM, AutoTokenizer
olmo = AutoModelForCausalLM.from_pretrained("allenai/Olmo-3-1025-7B")
tokenizer = AutoTokenizer.from_pretrained("allenai/Olmo-3-1025-7B")
message = ["Language modeling is "]
inputs = tokenizer(message, return_tensors='pt', return_token_type_ids=False)
# optional verifying cuda
# inputs = {k: v.to('cuda') for k,v in inputs.items()}
# olmo = olmo.to('cuda')
response = olmo.generate(**inputs, max_new_tokens=100, do_sample=True, top_k=0, temperature=1.0, top_p=0.7)
print(tokenizer.batch_decode(response, skip_special_tokens=True)[0])
>> 'Language modeling is a key component of any text-based application, but its effectiveness...'
For faster performance, you can quantize the model using the following method:
AutoModelForCausalLM.from_pretrained("allenai/Olmo-3-1025-7B",
torch_dtype=torch.float16,
load_in_8bit=True) # Requires bitsandbytes
The quantized model is more sensitive to data types and CUDA operations. To avoid potential issues, it's recommended to pass the inputs directly to CUDA using:
inputs.input_ids.to('cuda')
We have released checkpoints for these models. For pretraining, the naming convention is stage1-stepXXX. The conventions for midtraining and long context are stage2-stepXXX and stage3-stepXXX, respectively.
To load a specific model revision with HuggingFace, simply add the argument revision:
olmo = AutoModelForCausalLM.from_pretrained("allenai/Olmo-3-1025-7B", revision="stage1-step10000")
Or, you can access all the revisions for the models via the following code snippet:
from huggingface_hub import list_repo_refs
out = list_repo_refs("allenai/Olmo-3-1025-7B")
branches = [b.name for b in out.branches]
Fine-tuning
Model fine-tuning can be done from the final checkpoint (the main revision of this model) or many intermediate checkpoints. Two recipes for tuning are available.
- Fine-tune with the OLMo-core repository:
torchrun --nproc-per-node=8 ./src/scripts/official/OLMo3/OLMo-3-1025-7B-pretrain-1.py run01
You can override most configuration options from the command-line. For example, to override the learning rate you could launch the script like this:
torchrun --nproc-per-node=8 ./src/scripts/official/OLMo3/OLMo-3-1025-7B-pretrain-1.py run01 --train_module.optim.lr=3e-4
For more documentation, see the GitHub readme.
Model Description
- Developed by: Allen Institute for AI (Ai2)
- Model type: a Transformer style autoregressive language model.
- Language(s) (NLP): English
- License: The code and model are released under Apache 2.0.
- Contact: Technical inquiries:
olmo@allenai.org. Press:press@allenai.org - Date cutoff: Dec 2024
Model Sources
- Project Page: https://allenai.org/olmo
- Repositories:
- Core repo (training, inference, fine-tuning etc.): https://github.com/allenai/OLMo-core
- Evaluation code: https://github.com/allenai/OLMo-Eval
- Further fine-tuning code: https://github.com/allenai/open-instruct
- W&B Report: https://wandb.ai/ai2-llm/Olmo-3-1025-7B/reports/Olmo-3-7B-October-2025--VmlldzoxNDcwOTM0NA
- Paper: https://allenai.org/papers/olmo3
Evaluation
Core model results for MODELS are found below.
| Model | Olmo 3-Eval Math | BigCodeBench | HumanEval | DeepSeek LeetCode | DS 1000 | MBPP | MultiPL HumanEval | MultiPL MBPPP | Olmo 3-Eval Code | ARC MC | MMLU STEM | MedMCQA MC | MedQA MC | SciQ MC | Olmo 3-Eval MC_STEM | MMLU Humanities | MMLU Social Sci. | MMLU Other | CSQA MC | PIQA MC | SocialIQA MC | CoQA Gen2MC MC | DROP Gen2MC MC | Jeopardy Gen2MC MC | NaturalQs Gen2MC MC | SQuAD Gen2MC MC | Olmo 3-Eval MC_Non-STEM | HellaSwag RC | Winogrande RC | Lambada | Basic Skills | DROP | Jeopardy | NaturalQs | SQuAD | CoQA | Olmo 3-Eval GenQA | BBH | MMLU Pro MC | Deepmind Math | LBPP |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Open-weight Models | |||||||||||||||||||||||||||||||||||||||||
| Marin-8B | 39.6 | 21.5 | 31.6 | 0.5 | 16.5 | 36.5 | 15.6 | 27.6 | 21.4 | 89.2 | 58.1 | 52.7 | 47.3 | 93.2 | 68.1 | 71.4 | 77.4 | 68.3 | 75.3 | 85.7 | 79.8 | 86.2 | 63.7 | 90.8 | 71.5 | 96.5 | 78.8 | 84.0 | 88.6 | 73.9 | 85.6 | 73.0 | 72.7 | 42.6 | 93.4 | 69.5 | 75.9 | 55.6 | 38.8 | 20.2 | 5.8 |
| Apertus-8B | 29.2 | 20.9 | 21.6 | 0.6 | 11.8 | 33.5 | 15.5 | 29.2 | 19.0 | 87.9 | 52.4 | 51.7 | 47.6 | 91.9 | 66.3 | 67.8 | 74.7 | 66.1 | 72.1 | 80.5 | 76.3 | 82.8 | 47.5 | 90.3 | 66.7 | 91.3 | 74.2 | 81.0 | 85.8 | 70.9 | 83.8 | 37.1 | 70.1 | 35.0 | 89.6 | 67.4 | 69.0 | 48.1 | 33.9 | 17.1 | 7.1 |
| OLMo 2-7B | 41.7 | 8.8 | 16.3 | 0.2 | 10.1 | 21.2 | 4.2 | 12.2 | 10.4 | 85.7 | 53.2 | 49.2 | 43.8 | 90.9 | 64.6 | 67.9 | 73.1 | 65.2 | 72.0 | 80.1 | 77.5 | 85.0 | 55.6 | 89.5 | 66.3 | 95.3 | 75.2 | 82.2 | 87.4 | 70.5 | 82.2 | 61.5 | 70.8 | 37.4 | 91.5 | 68.3 | 72.4 | 49.6 | 33.1 | 16.3 | 3.1 |
| Qwen3-8B | 67.2 | 42.5 | 71.7 | 8.3 | 33.1 | 66.2 | 52.3 | 48.4 | 46.1 | 95.4 | 76.7 | 63.5 | 62.1 | 96.1 | 78.8 | 78.6 | 84.8 | 76.8 | 84.1 | 89.9 | 83.3 | 93.7 | 78.3 | 92.3 | 74.1 | 97.5 | 84.8 | 80.5 | 86.4 | 73.0 | 93.5 | 57.2 | 65.1 | 33.8 | 89.2 | 61.6 | 71.1 | 76.5 | 50.3 | 47.7 | 25.7 |
| Nemotron MiniD 8B | 49.8 | 43.2 | 71.7 | 6.8 | 30.3 | 62.3 | 40.0 | 47.5 | 43.1 | 94.1 | 71.1 | 54.5 | 53.5 | 94.3 | 73.5 | 78.0 | 82.2 | 73.8 | 74.4 | 86.0 | 78.7 | 92.2 | 70.0 | 90.7 | 71.1 | 97.4 | 81.3 | 80.2 | 86.2 | 67.9 | 91.4 | 71.4 | 64.9 | 31.2 | 92.3 | 60.4 | 71.8 | 77.0 | 50.2 | 31.4 | 31.7 |
| Gemma-2-9B | 48.8 | 30.9 | 40.0 | 1.9 | 28.4 | 49.1 | 27.9 | 38.2 | 30.2 | 92.7 | 62.8 | 58.9 | 55.4 | 94.4 | 72.8 | 74.5 | 82.9 | 74.2 | 75.3 | 85.7 | 80.3 | 92.7 | 65.8 | 92.8 | 72.5 | 97.3 | 81.3 | 81.8 | 88.8 | 76.3 | 89.3 | 68.2 | 75.1 | 40.4 | 88.8 | 71.5 | 75.6 | 68.8 | 44.7 | 23.0 | 12.4 |
| Qwen-2.5-7B | 60.7 | 39.7 | 66.1 | 5.1 | 35.2 | 55.4 | 40.3 | 45.4 | 41.0 | 93.4 | 67.6 | 60.3 | 56.6 | 95.4 | 74.7 | 76.2 | 83.0 | 74.4 | 85.0 | 88.5 | 82.9 | 93.5 | 69.1 | 92.1 | 70.5 | 96.4 | 82.9 | 81.0 | 86.0 | 70.3 | 91.4 | 56.7 | 63.0 | 31.2 | 87.0 | 40.5 | 67.5 | 54.7 | 48.1 | 32.8 | 22.1 |
| Llama-3.1-8B | 36.9 | 30.7 | 40.4 | 0.1 | 22.2 | 12.1 | 14.5 | 28.3 | 21.2 | 86.4 | 55.7 | 56.5 | 53.7 | 92.7 | 69.0 | 70.1 | 75.5 | 69.1 | 72.9 | 78.3 | 77.0 | 89.9 | 53.3 | 88.9 | 68.0 | 94.4 | 76.1 | 81.5 | 87.3 | 75.5 | 88.0 | 59.5 | 70.9 | 36.7 | 89.2 | 69.0 | 73.1 | 63.0 | 37.4 | 24.1 | 9.1 |
| Granite-3.3-8B | 41.5 | 0.4 | 0.0 | 0.0 | 22.6 | 48.5 | 22.3 | 32.3 | 18.0 | 86.2 | 55.6 | 49.6 | 43.0 | 90.8 | 65.0 | 67.6 | 71.8 | 64.5 | 82.3 | 81.5 | 83.1 | 87.6 | 55.0 | 88.4 | 69.2 | 94.5 | 76.9 | 83.7 | 89.4 | 76.0 | 88.7 | 38.4 | 69.7 | 37.0 | 89.6 | 37.8 | 67.8 | 61.5 | 33.9 | 32.2 | 18.5 |
| MiMo-7B | 54.3 | 38.3 | 57.0 | 1.2 | 28.1 | 48.3 | 34.5 | 42.5 | 35.7 | 91.7 | 63.5 | 56.2 | 53.0 | 93.5 | 71.6 | 73.6 | 80.8 | 72.7 | 76.1 | 87.2 | 80.7 | 91.4 | 64.1 | 89.5 | 72.2 | 96.7 | 80.5 | 80.6 | 86.5 | 73.1 | 89.7 | 69.3 | 65.6 | 33.1 | 90.3 | 54.4 | 71.4 | 75.1 | 44.3 | 25.4 | 21.5 |
| Olmo 3 7B | 54.7 | 34.1 | 49.1 | 1.4 | 20.2 | 43.6 | 28.7 | 38.2 | 30.7 | 89.2 | 59.7 | 48.3 | 41.8 | 92.8 | 66.4 | 68.9 | 75.0 | 66.9 | 75.3 | 80.2 | 80.3 | 92.5 | 67.3 | 86.9 | 69.4 | 96.9 | 78.2 | 77.7 | 85.7 | 68.9 | 89.5 | 71.5 | 60.4 | 32.6 | 93.5 | 72.8 | 72.5 | 63.5 | 37.3 | 23.7 | 17.1 |
Model Details
Stage 1: Initial Pretraining
- Dataset: dolma3_6T-mix-1025
- 5.93T tokens
- Coverage: 97.53%+ of total pretraining budget
Stage 2: Mid-training
- Dataset: dolma3-dolmino-mix-1025
- 100B tokens
- Mix composition: 20% code, 28% web pages, 19% math, 14% QA, 8% thinking, 6% instruction, and 5% PDFs
Stage 3: Long Context
- Dataset: dolma3-longmino-mix-1025
- 50B tokens
- Mix composition: 66% midtraining data, 34% PDFs
Model Merging
- 7B Model: No merging
- 32B Model: 2 versions on 100B mix, merged before starting long context run. Final checkpoint is merged 4 final checkpoints.
Bias, Risks, and Limitations
Like any base language model or fine-tuned model without safety filtering, these models can easily be prompted by users to generate harmful and sensitive content. Such content may also be produced unintentionally, especially in cases involving bias, so we recommend that users consider the risks when applying this technology. Additionally, many statements from OLMo or any LLM are often inaccurate, so facts should be verified.
License
This model is licensed under Apache 2.0. It is intended for research and educational use in accordance with Ai2's Responsible Use Guidelines.
Citation
A technical manuscript is forthcoming! Find the paper at: https://allenai.org/papers/olmo3
Model Card Contact
For errors in this model card, contact olmo@allenai.org.
🚀 If you find these models useful
Help me test my AI-Powered Quantum Network Monitor Assistant with quantum-ready security checks:
The full Open Source Code for the Quantum Network Monitor Service available at my github repos ( repos with NetworkMonitor in the name) : Source Code Quantum Network Monitor. You will also find the code I use to quantize the models if you want to do it yourself GGUFModelBuilder
💬 How to test:
Choose an AI assistant type:
TurboLLM(GPT-4.1-mini)HugLLM(Hugginface Open-source models)TestLLM(Experimental CPU-only)
What I’m Testing
I’m pushing the limits of small open-source models for AI network monitoring, specifically:
- Function calling against live network services
- How small can a model go while still handling:
- Automated Nmap security scans
- Quantum-readiness checks
- Network Monitoring tasks
🟡 TestLLM – Current experimental model (llama.cpp on 2 CPU threads on huggingface docker space):
- ✅ Zero-configuration setup
- ⏳ 30s load time (slow inference but no API costs) . No token limited as the cost is low.
- 🔧 Help wanted! If you’re into edge-device AI, let’s collaborate!
Other Assistants
🟢 TurboLLM – Uses gpt-4.1-mini :
- **It performs very well but unfortunatly OpenAI charges per token. For this reason tokens usage is limited.
- Create custom cmd processors to run .net code on Quantum Network Monitor Agents
- Real-time network diagnostics and monitoring
- Security Audits
- Penetration testing (Nmap/Metasploit)
🔵 HugLLM – Latest Open-source models:
- 🌐 Runs on Hugging Face Inference API. Performs pretty well using the lastest models hosted on Novita.
💡 Example commands you could test:
"Give me info on my websites SSL certificate""Check if my server is using quantum safe encyption for communication""Run a comprehensive security audit on my server"- '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code on. This is a very flexible and powerful feature. Use with caution!
Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is open source. Feel free to use whatever you find helpful.
If you appreciate the work, please consider buying me a coffee ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
- Downloads last month
- 1,232