---
license: apache-2.0
base_model: Qwen/Qwen3-0.6B
tags:
- legal
- sft
- lora
- trl
- eula
datasets:
- AxelDlv00/EULAI
language:
- fr
pipeline_tag: text-generation
---

<div align="center">
  <img src="icons/icon-base.png" alt="EULAI Logo" width="120">
  
  **You lie? EULAI!**
  
  **Local AI Browser Assistant for Legal Document Analysis**
  
  *[Axel Delaval](https://axeldlv00.github.io/axel-delaval-personal-page/) • 28 January 2026*
  <br />

  [![GitHub](https://img.shields.io/badge/Source_Code-GitHub-181717?style=for-the-badge&logo=github&logoColor=white)](https://github.com/AxelDlv00/EULAI)
[![License](https://img.shields.io/badge/LICENSE-MIT-yellow?style=for-the-badge)](./LICENSE) [![HF Model](https://img.shields.io/badge/%F0%9F%A4%97%20Model-EULAI--Base-green?style=for-the-badge)](https://huggingface.co/AxelDlv00/EULAI)[![HF MLC Model](https://img.shields.io/badge/%F0%9F%A4%97%20MLC_Weights-EULAI--q4f16-blue?style=for-the-badge)](https://huggingface.co/AxelDlv00/EULAI-q4f16_1-MLC) [![HF Dataset](https://img.shields.io/badge/%F0%9F%A4%97%20Dataset-EULAI-8A2BE2?style=for-the-badge)](https://huggingface.co/datasets/AxelDlv00/EULAI)

</div>

# EULAI - Fine-tuned Qwen3-0.6B for EULA Analysis

This model is a fine-tuned version of **Qwen/Qwen3-0.6B** using the **LoRA (Low-Rank Adaptation)** method. It specializes in analyzing contracts (EULAs) and privacy policies to extract structured risk-based summaries.

## Training Configuration

Details extracted from the training configuration:
- **Base Model**: `Qwen/Qwen3-0.6B`
- **Method**: PEFT / LoRA
- **LoRA Hyperparameters**:
  - `r`: 16
  - `alpha`: 32
  - `dropout`: 0.05
  - `target_modules`: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
- **Training**:
  - `epochs`: 3
  - `learning_rate`: 1e-4
  - `max_length`: 2048
  - `packing`: True

## Usage (Inference)

To use this model, you need to load the adapter on top of the original base model.

```python
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "Qwen/Qwen3-0.6B"
adapter_id = "AxelDlv00/EULAI"

tokenizer = AutoTokenizer.from_pretrained(adapter_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.bfloat16, device_map="auto")
model = PeftModel.from_pretrained(model, adapter_id)

prompt = "We collect your GPS data continuously even when the application is closed."
messages = [{"role": "user", "content": prompt}]

inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, enable_thinking=False, return_tensors="pt").to("cuda")

outputs = model.generate(inputs, max_new_tokens=512, temperature=0.1)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```

## Output Format

The model generates structured bullet points in this format:

```
- [BLOCKER/BAD/GOOD/NEUTRAL] : Title : Explanation
```