finetune_config.json

{
  "model_path": "/opt/moss/models/MOSS-TTS-Nano",
  "codec_path": "/opt/moss/models/MOSS-Audio-Tokenizer-Nano",
  "train_jsonl": "/tmp/train_with_codes.jsonl",
  "output_dir": "/vol/moss_ft/ckpt_qwen35h",
  "max_length": 1024,
  "per_device_batch_size": 4,
  "gradient_accumulation_steps": 8,
  "learning_rate": 1e-05,
  "weight_decay": 0.1,
  "adam_beta1": 0.9,
  "adam_beta2": 0.95,
  "adam_eps": 1e-08,
  "warmup_steps": 0,
  "warmup_ratio": 0.03,
  "lr_scheduler_type": "linear",
  "num_epochs": 3,
  "max_train_steps": null,
  "max_grad_norm": 1.0,
  "logging_steps": 10,
  "save_every_epochs": 1,
  "num_workers": 0,
  "mixed_precision": "bf16",
  "attn_implementation": "sdpa",
  "channelwise_loss_weight": "1,32",
  "seed": 42,
  "resolved_warmup_steps": 73,
  "resolved_channelwise_loss_weight": [
    1.0,
    2.0,
    2.0,
    2.0,
    2.0,
    2.0,
    2.0,
    2.0,
    2.0,
    2.0,
    2.0,
    2.0,
    2.0,
    2.0,
    2.0,
    2.0,
    2.0
  ],
  "global_batch_size": 32,
  "records_paths": [
    "/tmp/train_with_codes.jsonl"
  ],
  "saved_global_step": 2427,
  "saved_epoch": 3,
  "saved_at": "2026-06-11 17:00:04",
  "checkpoint_dir": "/vol/moss_ft/ckpt_qwen35h/checkpoint-last"
}