| { | |
| "model_config": { | |
| "name": "HunyuanDense", | |
| "model_path": "/", | |
| "trust_remote_code": true, | |
| "torch_dtype": "auto", | |
| "device_map": "cpu", | |
| "low_cpu_mem_usage": true, | |
| "use_cache": false | |
| }, | |
| "compression_config": { | |
| "name": "PTQ", | |
| "quantization": { | |
| "name": "int4_gptq", | |
| "bits": 4, | |
| "quant_method": { | |
| "weight": "per-group", | |
| "group_size": 128 | |
| }, | |
| "quant_helpers": [], | |
| "smooth_alpha": 0.5, | |
| "low_memory": false, | |
| "modules_to_quantize": [], | |
| "zero_point": true, | |
| "mse_range": false, | |
| "ignore_layers": [ | |
| "lm_head", | |
| "model.embed_tokens", | |
| "mlp.gate.wg" | |
| ], | |
| "quant_analyse": false | |
| } | |
| }, | |
| "dataset_config": { | |
| "name": "TextDataset", | |
| "data_path": "./dataset/PTQ_data.jsonl", | |
| "max_seq_length": 8192, | |
| "num_samples": 128, | |
| "batch_size": 1, | |
| "shuffle": false | |
| }, | |
| "global_config": { | |
| "save_path": "/", | |
| "max_seq_length": 8192, | |
| "hidden_size": 3072, | |
| "model_arch_type": "hunyuan_v1_dense", | |
| "deploy_backend": "vllm" | |
| }, | |
| "debug_info": { | |
| "python": "3.10.14 (main, Mar 6 2025, 18:16:47) [GCC 11.4.0]", | |
| "angelslim": { | |
| "name": "angelslim", | |
| "version": "be35620eca8ad4e6bf8263e84cee69d90c7045a4", | |
| "source": "git" | |
| }, | |
| "torch": { | |
| "name": "torch", | |
| "version": "2.7.0", | |
| "source": "pip" | |
| }, | |
| "transformers": { | |
| "name": "transformers", | |
| "version": "4.52.4", | |
| "source": "pip" | |
| }, | |
| "torch_cuda_version": "12.6" | |
| } | |
| } | |