adapter_config.json

{
    "adapter_path": "zen-nano/model",
    "alpha": [
        1e-05
    ],
    "batch_size": 2,
    "beta": 0.1,
    "config": null,
    "data": "zen-nano/data",
    "de_quantize": false,
    "delta": 50.0,
    "dpo_cpo_loss_type": "sigmoid",
    "epochs": null,
    "epsilon": 0.0001,
    "epsilon_high": null,
    "fuse": true,
    "grad_checkpoint": false,
    "gradient_accumulation_steps": 1,
    "group_size": 4,
    "grpo_loss_type": "grpo",
    "importance_sampling_level": null,
    "iters": 100,
    "judge": "mlx-community/Josiefied-Qwen2.5-7B-Instruct-abliterated-v2-4-bit",
    "judge_config": {},
    "learning_rate": 1e-05,
    "list_reward_functions": false,
    "load_in_4bits": false,
    "load_in_6bits": false,
    "load_in_8bits": false,
    "lora_parameters": {
        "rank": 8,
        "dropout": 0.0,
        "scale": 10.0
    },
    "lr_schedule": null,
    "mask_prompt": false,
    "max_completion_length": 512,
    "max_seq_length": 2048,
    "model": "mlx-community/Qwen3-4B-Instruct-2507-4bit",
    "num_layers": 16,
    "optimizer": "adam",
    "optimizer_config": {
        "adam": {},
        "adamw": {},
        "muon": {},
        "qhadam": {}
    },
    "reference_model_path": null,
    "resume_adapter_file": null,
    "reward_functions": null,
    "reward_functions_file": null,
    "reward_scaling": 1.0,
    "reward_weights": null,
    "save_every": 100,
    "seed": 0,
    "steps_per_eval": 200,
    "steps_per_report": 10,
    "temperature": 1.0,
    "test": false,
    "test_batches": 500,
    "train": true,
    "train_mode": "sft",
    "train_type": "lora",
    "val_batches": 25,
    "wandb": null
}