| { | |
| "architectures": [ | |
| "LlamaForCausalLM" | |
| ], | |
| "model_type": "llama", | |
| "hidden_size": 4096, | |
| "intermediate_size": 12288, | |
| "num_hidden_layers": 1, | |
| "num_attention_heads": 32, | |
| "num_key_value_heads": 8, | |
| "head_dim": 128, | |
| "vocab_size": 151936, | |
| "draft_vocab_size": 32000, | |
| "rope_theta": 1000000, | |
| "rms_norm_eps": 1e-06, | |
| "torch_dtype": "bfloat16", | |
| "tie_word_embeddings": false | |
| } |