TinyLlama_v1.1-qlora/TinyLlama_v1.1-qlora-logiqa-r32-a2/training_configuration.json

{
  "model_id": "TinyLlama/TinyLlama_v1.1",
  "dataset": {
    "name": "LOGIQA",
    "dataset_id": "data/logiqa_train",
    "preprocess_id": "logiqa_train_deepeval"
  },
  "peft_config": {
    "method": "qlora",
    "rank": 32,
    "alpha": 64,
    "dropout": 0.0,
    "bias": "none",
    "target_modules": [
      "q_proj",
      "k_proj",
      "v_proj",
      "o_proj",
      "gate_proj",
      "down_proj",
      "up_proj"
    ],
    "trainable_parameter_count": 25231360
  },
  "training_config": {
    "max_dataset_length": null,
    "batch_size": 64,
    "per_device_batch_size": 32,
    "gradient_accumulation_steps": 2,
    "learning_rate": 0.0003,
    "num_epochs": 3,
    "warmup_ratio": 0.1
  },
  "model_name": "TinyLlama_v1.1-qlora-logiqa-r32-a2",
  "output_dir": "./experiment_results/TinyLlama_v1.1-qlora/TinyLlama_v1.1-qlora-logiqa-r32-a2",
  "seed": 42,
  "timestamp": "2025-08-31T03:17:56.980068"
}