{
  "d_model": 256,
  "num_heads": 8,
  "num_encoder_layers": 4,
  "num_decoder_layers": 4,
  "d_ff": 1024,
  "dropout": 0.1,
  "label_smoothing": 0.1,
  "learning_rate": 0.0003,
  "warmup_steps": 2000,
  "batch_size": 64,
  "grad_clip_norm": 1.0,
  "max_seq_len": 512,
  "seed": 42,
  "epochs": 50,
  "patience": 5,
  "bpe_vocab_size": 8000,
  "positional_encoding": "sinusoidal",
  "attention_type": "mha",
  "norm_type": "rmsnorm",
  "tokenization": "subword",
  "gqa_kv_heads": 2,
  "blt_patch_size": 9,
  "d_local": 256,
  "num_local_layers": 4,
  "run_name": "c4",
  "num_params": 7881192,
  "device": "cuda"
}