README.md

# Fine-tuned Model: Judge-Adaptor-v3

## 📚 Training Configuration
- **data_path**: `QomSSLab/Legal_SyntheticDraftRuling_Selected_v2`
- **output_dir**: `gemma312b_lora_chckpnts`
- **new_model_name**: `Judge-Adaptor-v3`
- **data_ratio**: `1.0`
- **model_name**: `QomSSLab/Legal-gemma3-12b-it-lora-thinking`
- **use_4bit**: `False`
- **use_lora**: `True`
- **max_seq_length**: `40000`
- **batch_size**: `1`
- **gradient_accu**: `8`
- **epochs**: `2`
- **learning_rate**: `1e-05`
- **lora_alpha**: `64`
- **lora_drop**: `0.05`
- **lora_r**: `64`
- **tune_embedding_layer**: `False`
- **hf_token**: `********`
- **resume_from_checkpoint**: `False`
- **use_8bit_optimizer**: `True`
- **push_to_hub**: `True`
- **push_lora_only**: `True`
- **train_only_on_assistant**: `True`
- **last_response**: `False`

---
Auto-generated after training.