Skip to content

llmcompressor.entrypoints.model_free.save_utils

Functions:

  • update_config –

    Write the final quantization config to config.json in save_directory.

update_config 

update_config(
    save_directory: str | PathLike,
    config: QuantizationConfig,
    converter: Converter | None = None,
) -> None

Write the final quantization config to config.json in save_directory.

Reads any pre-existing CT quantization_config from the checkpoint so that sequential ptq runs append rather than overwrite. Chains through the user converter (if any) then ModelFreePtqConverter to produce the final config.

Source code in src/llmcompressor/entrypoints/model_free/save_utils.py 
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
def update_config(
    save_directory: str | os.PathLike,
    config: QuantizationConfig,
    converter: Converter | None = None,
) -> None:
    """
    Write the final quantization config to config.json in save_directory.

    Reads any pre-existing CT quantization_config from the checkpoint so that
    sequential ptq runs append rather than overwrite. Chains through the user
    converter (if any) then ModelFreePtqConverter to produce the final config.
    """
    config_file_path = find_config_path(save_directory)

    # Read config.json once; seed incoming from any pre-existing CT quant config
    config_data: dict | None = None
    incoming: QuantizationConfig | None = None
    if config_file_path is not None:
        with open(config_file_path, "r") as file:
            config_data = json.load(file)
        if QUANTIZATION_CONFIG_NAME in config_data:
            qdata = config_data[QUANTIZATION_CONFIG_NAME]
            qdata.pop(COMPRESSION_VERSION_NAME, None)
            try:
                incoming = QuantizationConfig.model_validate(qdata)
            except ValidationError:
                incoming = None

    # Chain: user converter (if any) then mfptq
    if converter is not None and hasattr(converter, "update_config"):
        incoming = converter.update_config(incoming)
    mfptq = ModelFreePtqConverter(config)
    final_config = mfptq.update_config(incoming)

    qconfig_data = final_config.model_dump()
    qconfig_data = {
        COMPRESSION_VERSION_NAME: ct_version,
        QUANTIZATION_METHOD_NAME: "compressed-tensors",
        TRANSFORM_CONFIG_NAME: {},
        **qconfig_data,
    }

    if config_data is not None:
        config_data[QUANTIZATION_CONFIG_NAME] = qconfig_data
        with open(config_file_path, "w") as file:
            json.dump(config_data, file, indent=2, sort_keys=True)
    else:
        logger.warning(
            f"Could not find config file in {save_directory}. Please set "
            "quantization_config to: \n"
            f"{json.dumps(qconfig_data, indent=2, sort_keys=True)}"
        )