1{ 2 "architectures": [ 3 "LlamaForCausalLM" 4 ], 5 "bos_token_id": 128000, 6 "eos_token_id": 128001, 7 "head_dim": 64, 8 "hidden_size": 2048, 9 "initializer_range": 0.02, 10 "intermediate_size": 8192, 11 "max_position_embeddings": 131072, 12 "model_type": "llama", 13 "num_attention_heads": 32, 14 "num_hidden_layers": 16, 15 "num_key_value_heads": 8, 16 "rms_norm_eps": 1e-05, 17 "rope_theta": 500000.0, 18 "tie_word_embeddings": true, 19 "torch_dtype": "bfloat16", 20 "transformers_version": "4.45.0.dev0", 21 "vocab_size": 128256, 22 "tokenizer": "pretrained_fast" 23} 24