|
|
--- |
|
|
license: other |
|
|
license_url: https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/LICENSE |
|
|
tags: |
|
|
- llama3 |
|
|
- instruction-tuning |
|
|
- summarization |
|
|
- fine-tuned |
|
|
- merged |
|
|
--- |
|
|
|
|
|
# 🧠 FlamingNeuron / llama381binstruct_summarize_short_merged |
|
|
|
|
|
This is a **merged model** based on [NousResearch/Meta-Llama-3.1-8B-Instruct](https://huggingface.co/NousResearch/Meta-Llama-3.1-8B-Instruct), fine-tuned using LoRA adapters for legal-domain summarization. The LoRA weights have been merged with the base model for standalone use. |
|
|
|
|
|
## 🔍 Task |
|
|
|
|
|
This model converts legalese into short, human-readable summaries, based on data from the [legal_summarization](https://github.com/lauramanor/legal_summarization) project. |
|
|
|
|
|
## 💡 Example Usage |
|
|
|
|
|
For complete setup instructions and working inference examples, see: |
|
|
👉 [GitHub Repo: LLaMA3-demo](https://github.com/BQ31X/LLaMA3-demo) |
|
|
|
|
|
[](https://colab.research.google.com/github/BQ31X/LLaMA3-demo/blob/main/FlamingNeuron_ModelTest_20250418.ipynb) |
|
|
|
|
|
|
|
|
This model expects Meta-style structured prompts with two fields: `original_text` and `reference_summary`. |
|
|
The `original_text` contains the input passage, and the model generates a summary in place of the empty `reference_summary`. |
|
|
|
|
|
|
|
|
## 🏋️ Training Procedure |
|
|
|
|
|
This model was trained using **Supervised Fine-Tuning (SFT)** on legal document summaries using the [legal_summarization](https://github.com/lauramanor/legal_summarization) dataset. LoRA adapters were applied during training and merged afterward using `merge_and_unload()`. |
|
|
|
|
|
### ⚙️ Framework Versions |
|
|
|
|
|
- TRL: 0.16.1 |
|
|
- Transformers: 4.51.3 |
|
|
- PyTorch: 2.6.0+cu124 |
|
|
- Datasets: 3.5.0 |
|
|
- Tokenizers: 0.21.1 |
|
|
|
|
|
## 📚 Citations |
|
|
|
|
|
This model was fine-tuned using [TRL](https://github.com/huggingface/trl). |
|
|
|
|
|
## ⚖️ Legal Notice |
|
|
|
|
|
This model builds on Meta’s LLaMA 3.1 architecture and is governed by the [LLaMA 3.1 Community License](https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/LICENSE). All use must comply with Meta’s [acceptable use policy](https://ai.meta.com/llama/use-policy/). |
|
|
|
|
|
It was fine-tuned using the [legal_summarization dataset](https://github.com/lauramanor/legal_summarization) for research and educational purposes only. |
|
|
|
|
|
This model is not intended for commercial use exceeding the limitations described in the Meta license (e.g. more than 700M monthly active users). |
|
|
|
