smash_config.json

{
    "batchers": null,
    "cachers": null,
    "compilers": null,
    "distillers": null,
    "pruners": null,
    "quantizers": "llm-int8",
    "recoverers": null,
    "quant_llm-int8_compute_dtype": "bfloat16",
    "quant_llm-int8_double_quant": false,
    "quant_llm-int8_enable_fp32_cpu_offload": false,
    "quant_llm-int8_has_fp16_weight": false,
    "quant_llm-int8_quant_type": "fp4",
    "quant_llm-int8_threshold": 6.0,
    "quant_llm-int8_weight_bits": 8,
    "max_batch_size": 1,
    "device": "cuda",
    "cache_dir": "/tmp/models/tmpuqkqj4qo",
    "task": "",
    "save_load_fn": "llm-int8",
    "save_load_fn_args": {},
    "api_key": null
}