{
  "architectures": [
    "Qwen2ForCausalLM"
  ],
  "model_type": "qwen2",
  "vocab_size": 32000,
  "hidden_size": 384,
  "intermediate_size": 1024,
  "num_hidden_layers": 8,
  "num_attention_heads": 6,
  "num_key_value_heads": 6,
  "head_dim": 64,
  "max_position_embeddings": 128,
  "rms_norm_eps": 1e-06,
  "rope_theta": 10000.0,
  "rope_scaling": {},
  "hidden_act": "silu",
  "attention_dropout": 0.0,
  "attention_bias": false,
  "tie_word_embeddings": false,
  "torch_dtype": "float32",
  "dtype": "float32",
  "use_cache": true,
  "use_sliding_window": false,
  "sliding_window": null,
  "bos_token_id": 0,
  "eos_token_id": 2,
  "pad_token_id": 7,
  "unk_token_id": 9,
  "mask_token_id": 8,
  "_sharf": {
    "name": "SHARF Santa Maria 30M",
    "arch": "Decoder-only · Pre-norm · RMSNorm(fp32) · RoPE(split-half, Qwen2/LLaMA) · MHA · SwiGLU · SDPA",
    "framework": "torch 2.9.0 · torch_xla 2.9.0",
    "best_val": 3.381039547920227,
    "ppl": 29.40131931272329,
    "step": 20000,
    "epoch": 2,
    "exported_at": "2026-10-18 04:20:29"
  }
}