{ "_name_or_path": "microsoft/phi-4", "architectures": ["AutoModelForCausalLM"], "model_type": "causal_lm", "num_hidden_layers": 32, "num_attention_heads": 16, "num_key_value_heads": 8, "hidden_size": 4096, "vocab_size": 50257, "torch_dtype": "bfloat16", "tie_word_embeddings": false }