{
  "architectures": [
    "MarkForCausalLM"
  ],
  "model_type": "mark",
  "model_name": "Mark Engineer 300M",
  "model_version": "1.0.0",
  "total_parameters": 302019848,
  "active_parameters_per_token": "110M (padrao) a 165M (top-4)",
  "vocab_size": 4311,
  "hidden_size": 1024,
  "intermediate_size": 3072,
  "num_hidden_layers": 8,
  "num_attention_heads": 16,
  "num_key_value_heads": 4,
  "use_mla": true,
  "q_lora_rank": 512,
  "kv_lora_rank": 256,
  "qk_rope_head_dim": 32,
  "v_head_dim": 64,
  "n_routed_experts": 16,
  "n_shared_experts": 1,
  "num_experts_per_tok": 2,
  "moe_intermediate_size": 1536,
  "recursive_depth_loops": 2,
  "mtp_depth": 1,
  "max_position_embeddings": 4096,
  "rms_norm_eps": 1e-05,
  "rope_theta": 500000.0,
  "tie_word_embeddings": true,
  "torch_dtype": "bfloat16",
  "quantization": "IQ4_NL",
  "download": {
    "gguf_url": "https://github.com/thu12dta3/Mark-AI/releases/download/v1.0.0-302m/mark_engineer_302m.gguf",
    "pt_url": "https://github.com/thu12dta3/Mark-AI/releases/download/v1.0.0-302m/mark_engineer_302m_iq4_nl.pt",
    "sha256": "daf05dd0dab4fc90f620d959cb6e5731c49b3f3c4ddce0ff28765e0751075fd9"
  }
}