part2_harmful_train_metadata.json

{
  "base_model_id": "Qwen/Qwen2.5-1.5B-Instruct",
  "dataset_id": "unalignment/toxic-dpo-v0.2",
  "train_rows": 486,
  "val_rows": 55,
  "max_length": 768,
  "learning_rate": 0.0001,
  "num_epochs": 3,
  "train_batch_size": 4,
  "gradient_accumulation_steps": 16,
  "effective_batch_size": 64,
  "lora_rank": 16,
  "lora_alpha": 32,
  "lora_dropout": 0.05,
  "lora_target_modules": [
    "q_proj",
    "k_proj",
    "v_proj",
    "o_proj",
    "gate_proj",
    "up_proj",
    "down_proj"
  ],
  "adapter_dir": "/root/SafeGenAI/work/models/model_harmful_lora_adapter",
  "train_runtime_seconds": 116.0994
}