args.json

{
  "batch_size": 1,
  "accum_iter": 8,
  "epochs": 100,
  "warmup_epochs": 0.001,
  "lr": 0.0002,
  "min_lr": 0.0002,
  "wd": 0.1,
  "clip_grad": 4.0,
  "init_from": "/mnt/petrelfs/gaopeng/synbol/mgpt-dev-main-7B-768/lumina_mgpt/output/7B_2e-4_1_10240_768_64A100_final_stage/epoch0-iter20999",
  "data_config": "/mnt/petrelfs/gaopeng/synbol/mgpt-dev-main-7B-768/lumina_mgpt/configs/data/sample.yaml",
  "cache_ann_on_disk": true,
  "length_clustering": true,
  "num_workers": 8,
  "pin_mem": true,
  "seed": 0,
  "output_dir": "output/7B_2e-4_1_10240_768_64A100_final_stage",
  "save_interval": 1,
  "save_iteration_interval": 1000,
  "only_save_trainable": false,
  "ckpt_max_keep": 2,
  "auto_resume": true,
  "resume_path": "output/7B_2e-4_1_10240_768_64A100_final_stage/epoch0-iter20999",
  "model_parallel_size": 1,
  "data_parallel": "fsdp",
  "precision": "bf16",
  "grad_precision": "fp32",
  "checkpointing": true,
  "max_seq_len": 10240,
  "mask_image_logits": false,
  "dropout": 0.05,
  "z_loss_weight": 1e-05,
  "model_size": "7B",
  "world_size": 64,
  "rank": 0,
  "gpu": 0,
  "local_rank": 0,
  "dist_url": "env://",
  "distributed": true,
  "dist_backend": "nccl"
}