adapter_config.json

{
    "adapter_path": "/Users/matias.vizcaino/Documents/datagero_repos/llms4ol-2026/data/ft/adapters_9b_full/seg_04",
    "batch_size": 1,
    "clear_cache_threshold": 1,
    "config": null,
    "data": "/Users/matias.vizcaino/Documents/datagero_repos/llms4ol-2026/data/ft/mlx35_nothink_full",
    "fine_tune_type": "lora",
    "grad_accumulation_steps": 4,
    "grad_checkpoint": true,
    "iters": 950,
    "learning_rate": 7e-06,
    "lora_parameters": {
        "rank": 8,
        "dropout": 0.0,
        "scale": 20.0
    },
    "lr_schedule": null,
    "mask_prompt": true,
    "max_seq_length": 1024,
    "model": "mlx-community/Qwen3.5-9B-4bit",
    "num_layers": 8,
    "optimizer": "adamw",
    "optimizer_config": {
        "adam": {},
        "adamw": {},
        "muon": {},
        "sgd": {},
        "adafactor": {}
    },
    "project_name": null,
    "report_to": null,
    "resume_adapter_file": "/Users/matias.vizcaino/Documents/datagero_repos/llms4ol-2026/data/ft/adapters_9b_full/seg_03/0000800_adapters.safetensors",
    "save_every": 50,
    "seed": 46,
    "steps_per_eval": 400,
    "steps_per_report": 25,
    "test": false,
    "test_batches": 500,
    "train": true,
    "val_batches": 10
}