| { | |
| "architectures": [ | |
| "LlamaForCausalLM" | |
| ], | |
| "model_type": "llama", | |
| "hidden_size": 4096, | |
| "intermediate_size": 11008, | |
| "num_hidden_layers": 32, | |
| "num_attention_heads": 32, | |
| "max_position_embeddings": 4096, | |
| "rms_norm_eps": 1e-06, | |
| "vocab_size": 32000, | |
| "bos_token_id": 1, | |
| "eos_token_id": 2, | |
| "quantization_config": { | |
| "load_in_4bit": true, | |
| "bnb_4bit_compute_dtype": "float16", | |
| "bnb_4bit_quant_type": "nf4", | |
| "bnb_4bit_use_double_quant": true | |
| }, | |
| "fine_tuned": true, | |
| "base_model": "unsloth/llama-2-7b-bnb-4bit", | |
| "training_args": { | |
| "num_epochs": 2, | |
| "learning_rate": 2e-4, | |
| "max_seq_length": 700, | |
| "batch_size": 1, | |
| "gradient_accumulation_steps": 8 | |
| } | |
| } |