training_args.json

{
  "dataset": "/root/lang_mix.jsonl",
  "tokenizer": "/root/tokenizers",
  "out_dir": "/root/models/run_cen_V",
  "use_hf": false,
  "hf_name": null,
  "hf_subset": null,
  "hf_split": "train",
  "hf_streaming": true,
  "hf_text_field": "text",
  "hf_messages_field": "messages",
  "shuffle_buffer_size": 10000,
  "seed": 1337,
  "init_from_checkpoint": null,
  "finetune_from": null,
  "strict_vocab_match": false,
  "save_tag": null,
  "save_every_n": 2000,
  "keep_last_k": 3,
  "save_on_improve": false,
  "improve_delta": 0.0,
  "resume_from_checkpoint": null,
  "log_loss_to_csv": false,
  "dataset_text_field": "text",
  "min_sample_token_length": 8,
  "stream_local_dataset": false,
  "local_dataset_shuffle_buffer": 2048,
  "block_count_sample_fraction": 0.02,
  "block_count_min_sample_megabytes": 32,
  "block_count_max_sample_megabytes": 512,
  "precision": "bf16",
  "optimizer": "adamw",
  "lr": 0.0003,
  "weight_decay": 0.06,
  "beta2": 0.98,
  "adam_eps": 1e-08,
  "grad_clip": 0.8,
  "optim_eps": 1e-08,
  "lr_scheduler": "cosine",
  "min_lr_ratio": 0.05,
  "muon_lr": null,
  "muon_momentum": 0.95,
  "muon_exclude_embeddings": true,
  "muon_beta1": null,
  "muon_beta2": null,
  "muon_eps": null,
  "muon_bias_correction": true,
  "muon_clip_by_layer": false,
  "muon_lr_correction": true,
  "batch_size": 4,
  "accum_steps": 16,
  "epochs": 2,
  "warmup_frac": 0.05,
  "log_every_n": 10,
  "overfit_subset": null,
  "use_gradient_checkpoint": false,
  "num_workers": 2,
  "pin_memory": true,
  "persistent_workers": true,
  "sft_mode": "lora",
  "lora_r": 64,
  "lora_alpha": 96,
  "lora_dropout": 0.05,
  "include_agent_end": true,
  "include_eos": false,
  "mask_user_queries": true
}