{ "model_type": "deepseek", "architectures": [ "DeepSeekForCausalLM" ], "vocab_size": 50257, "hidden_size": 512, "num_hidden_layers": 6, "num_attention_heads": 8, "max_position_embeddings": 1024, "use_mla": true, "mla_kv_heads": 4, "moe_num_experts": 4, "multi_token_predict": 2, "pad_token_id": 0, "bos_token_id": 1, "eos_token_id": 2 }