smash_config.json

{
    "batcher": null,
    "cacher": null,
    "compiler": "torch_compile",
    "factorizer": null,
    "kernel": null,
    "pruner": null,
    "quantizer": "hqq",
    "hqq_backend": "torchao_int4",
    "hqq_compute_dtype": "torch.bfloat16",
    "hqq_force_hf_implementation": false,
    "hqq_group_size": 128,
    "hqq_use_torchao_kernels": true,
    "hqq_weight_bits": 4,
    "torch_compile_backend": "cudagraphs",
    "torch_compile_dynamic": true,
    "torch_compile_fullgraph": false,
    "torch_compile_make_portable": true,
    "torch_compile_max_kv_cache_size": 1600,
    "torch_compile_mode": "reduce-overhead",
    "torch_compile_seqlen_manual_cuda_graph": 800,
    "torch_compile_target": "model",
    "batch_size": 1,
    "device": "cuda:0",
    "device_map": null,
    "save_fns": [
        "hqq",
        "save_before_apply"
    ],
    "load_fns": [
        "torch_artifacts",
        "hqq"
    ],
    "reapply_after_load": {
        "factorizer": null,
        "pruner": null,
        "quantizer": null,
        "kernel": null,
        "cacher": null,
        "compiler": "torch_compile",
        "batcher": null
    }
}