training_report.json

{
  "model_base": "Qwen/Qwen3-4B-Instruct-2507",
  "model_name": "RegTech-4B-Instruct",
  "dataset": "./train.jsonl",
  "env_file": "/home/ubuntu/sophia-core-server/.tuning/.env.4B",
  "train_samples": 2330,
  "eval_samples": 258,
  "params": {
    "rank": 4,
    "alpha": 8,
    "dropout": 0.05,
    "lr": 1e-05,
    "scheduler": "cosine",
    "epochs": 1,
    "effective_batch": 8,
    "max_seq_length": 4096,
    "neftune_alpha": 0.0,
    "target_modules": [
      "q_proj",
      "k_proj",
      "v_proj",
      "o_proj",
      "gate_proj",
      "up_proj",
      "down_proj"
    ]
  },
  "results": {
    "total_steps": 292,
    "final_train_loss": 1.5045,
    "best_eval_loss": 1.601854681968689,
    "best_eval_step": 240,
    "best_token_accuracy": 0.6812,
    "elapsed_minutes": 8.6
  },
  "loss_history": {
    "train": [
      [
        10,
        2.1906
      ],
      [
        20,
        2.0417
      ],
      [
        30,
        2.1217
      ],
      [
        40,
        2.0513
      ],
      [
        50,
        1.9839
      ],
      [
        60,
        1.9423
      ],
      [
        70,
        1.9321
      ],
      [
        80,
        1.8047
      ],
      [
        90,
        1.7045
      ],
      [
        100,
        1.8603
      ],
      [
        110,
        1.721
      ],
      [
        120,
        1.6419
      ],
      [
        130,
        1.5821
      ],
      [
        140,
        1.5593
      ],
      [
        150,
        1.4756
      ],
      [
        160,
        1.4945
      ],
      [
        170,
        1.5168
      ],
      [
        180,
        1.5689
      ],
      [
        190,
        1.3763
      ],
      [
        200,
        1.5759
      ],
      [
        210,
        1.477
      ],
      [
        220,
        1.4889
      ],
      [
        230,
        1.4514
      ],
      [
        240,
        1.441
      ],
      [
        250,
        1.427
      ],
      [
        260,
        1.4423
      ],
      [
        270,
        1.4199
      ],
      [
        280,
        1.457
      ],
      [
        290,
        1.5045
      ]
    ],
    "eval": [
      [
        80,
        2.036996841430664
      ],
      [
        160,
        1.6603444814682007
      ],
      [
        240,
        1.601854681968689
      ]
    ],
    "token_accuracy": [
      [
        80,
        0.661
      ],
      [
        160,
        0.6759
      ],
      [
        240,
        0.6812
      ]
    ]
  }
}