sft_metadata.json

{
  "base_model": "DedeProGames/DynamicMind-Mini",
  "dataset": "HuggingFaceTB/smol-smoltalk",
  "dataset_config": "default",
  "dataset_split": "train",
  "training_type": "full_parameter_supervised_finetuning",
  "loss_mask": "assistant_content_and_eos_only",
  "global_step": 3440,
  "tokens_seen": 507248640,
  "supervised_tokens_seen": 361226420,
  "sequence_length": 1024,
  "effective_batch_sequences": 144,
  "tokens_per_step": 147456,
  "peak_learning_rate": 0.0002,
  "warmup_steps": 100,
  "lr_schedule": "constant_after_warmup",
  "optimizer": "AdamW",
  "betas": [
    0.9,
    0.95
  ],
  "weight_decay": 0.1,
  "grad_clip": 1.0,
  "seed": 1337,
  "final_loss": 1.917157530784607,
  "recipe_source": "BananaMind/BananaMind-2-Nano-Chat (adapted to 1024-token context)"
}