{
  "model_family": "Cosmos-T3",
  "model_name": "Cosmos-T3-Instruct",
  "model_class_name": "CosmosT2_Accelerate_LLM",
  "hf_repo_id": "wop/Cosmos-SFT",
  "tokenizer_name": "Qwen/Qwen2.5-0.5B",
  "dataset_name": "wop/smallestdata-sft",
  "dataset_split": "train",
  "dataset_row_limit": 19,
  "rows_used": 19,
  "stream_dataset": true,
  "shuffle_buffer_size": 10000,
  "train_val_fraction": 0.02,
  "seed": 42,
  "block_size": 1024,
  "max_len": 1024,
  "d_model": 768,
  "n_layers": 12,
  "n_heads": 12,
  "n_kv_heads": 4,
  "d_ff": 2048,
  "rope_base": 10000,
  "dropout": 0.0,
  "use_engram": true,
  "engram_every": 4,
  "engram_buckets": 8192,
  "engram_dim": 64,
  "engram_order": 3,
  "epochs": 6,
  "batch_size": 4,
  "lr": 0.0003,
  "weight_decay": 0.1,
  "warmup_steps": 10,
  "grad_clip": 1.0,
  "eval_every_steps": 2000,
  "val_max_batches": 20,
  "early_stopping_patience": 1000000,
  "num_workers": 0,
  "use_kv_cache": true,
  "train_steps_per_epoch": 1,
  "loss_tokens_seen": 1146,
  "samples_seen": 96,
  "config_hash": "f69d7063825b2fcf"
}