training_metadata.json

{
  "global_step": 40000,
  "epoch": 3,
  "train_config": {
    "lr": 1e-05,
    "weight_decay": 1e-05,
    "warmup_steps": 800,
    "max_grad_norm": 1.0,
    "train_llm": true,
    "llm_lr": 1e-05,
    "max_steps": 50000,
    "log_every_n_steps": 10,
    "val_every_n_steps": 5000,
    "generate_every_n_steps": 5000,
    "save_every_n_steps": 5000,
    "num_samples_to_generate": 16,
    "generation_max_new_tokens": 128,
    "generation_temperature": 0.4,
    "generation_do_sample": true,
    "compute_loss_on_assistant_only": true,
    "wandb_project": "smol-loravlm",
    "wandb_run_name": "base-ft-continue-3k"
  },
  "checkpoint_config": {
    "checkpoint_dir": "checkpoints/smol_loravlm_base_ft_from_3000",
    "push_to_hub": false,
    "hub_model_id": "toilaluan/ai0",
    "hub_token": null,
    "hub_private_repo": true
  },
  "model_config": {
    "text_ckpt": "google/embeddinggemma-300m",
    "image_ckpt": "google/siglip2-so400m-patch16-256",
    "llm_ckpt": "checkpoints/smol_loravlm_base_ft/checkpoint_step_3000/llm",
    "lora_rank": 16,
    "resume_from_checkpoint": "checkpoints/smol_loravlm_base_ft/checkpoint_step_3000",
    "lora_target": "qkvm",
    "prompt_aware": false,
    "lora_use_prefix": true
  }
}