| 1234567891011121314151617181920212223242526 |
- model:
- base_model: /path/to/Baichuan2-7B-Chat
- data:
- train_file: data/processed/multitask_train.jsonl
- lora:
- r: 8
- alpha: 32
- dropout: 0.05
- target_modules: [q_proj, k_proj, v_proj]
- train:
- output_dir: checkpoints/lora-multitask
- epochs: 3
- batch_size: 1
- gradient_accumulation_steps: 8
- learning_rate: 0.0002
- max_length: 4096
- multitask:
- task_weights:
- opinion_generation: 0.45
- result_generation: 0.55
- balance_by_difficulty: true
- difficulty_range: [1, 5]
|
