fix wrong match of ignore_layers and use warning instead of error for mismatch

[xin3he]

Description

layers.1 now matches layers.1x, which is not expected.

Type of Change

• Bug fix
• New feature
• Documentation update
• Performance improvement
• Code refactoring
• Other (please specify):

Related Issues

Fixes or relates to #

Checklist Before Submitting

• My code has been tested locally.
• Documentation has been updated as needed.
• New or updated tests are included where applicable.

[Copilot]

Pull request overview

This PR aims to fix overly-broad matching for ignore_layers (e.g., layers.1 unintentionally matching layers.1x), so that layer-exclusion behaves as users expect during quantization config expansion.

Changes:

• Normalizes ignore_layers earlier in set_layer_config (splitting by comma and appending . for digit-suffixed names).
• Changes unmatched layer_config entries from raising an error to logging a warning.
• Removes string normalization inside get_fp_layer_names.

Comments suppressed due to low confidence (1)

auto_round/compressors/utils.py:966

• get_fp_layer_names is now iterating over ignore_layers without normalizing when a string is passed. Since strings are iterable, callers passing the documented comma-separated string (e.g. existing tests) will iterate character-by-character and return incorrect results. Either restore string parsing inside this function (accept str | Sequence[str]) or update the signature/docstring and all call sites accordingly.

def get_fp_layer_names(model: torch.nn.Module, ignore_layers: str):
    """Identifies and returns layers in the model to exclude from quantization.

    This function processes a comma-separated list of fully precision (FP) layers,
    matches them to the names of layers in the model, and returns a list of such
    layers to exclude from quantization.

    Args:
        model (torch.nn.Module): The model whose layers will be inspected.
        ignore_layers (str): A comma-separated string of layer names to be excluded
            from quantization. Whitespace is ignored in this string.

    Returns:
        list: A list of layer names that match the specified FP layers or are
        subcomponents of those layers.
    """
    from auto_round.utils import SUPPORTED_LAYER_TYPES

    if not ignore_layers:
        return []

    all_layer_names = []
    for n, m in model.named_modules():
        if type(m) in SUPPORTED_LAYER_TYPES:
            all_layer_names.append(n)
    not_to_quantized_layers = []

    for fp_layer in ignore_layers:

---

You can also share your feedback on Copilot code review. Take the survey.