{ "model_type": "qwen2", "architectures": [ "Qwen2ForCausalLM" ], "hidden_size": 1536, "num_hidden_layers": 18, "torch_dtype": "bfloat16", "checkpoint_step": 500 }