huggingface/transformers · error · TypeError

Input weight should be of type nn.Parameter, got {type(weigh

Error message

Input weight should be of type nn.Parameter, got {type(weight)} instead

What it means

Error "Input weight should be of type nn.Parameter, got {type(weight)} instead" thrown in huggingface/transformers.

Source

Thrown at src/transformers/integrations/bitsandbytes.py:247

    if not has_been_replaced:
        logger.warning(
            "You are loading your model using eetq but no linear modules were found in your model."
            " Please double check your model architecture, or submit an issue on github if you think this is"
            " a bug."
        )
    return model


# Copied from PEFT: https://github.com/huggingface/peft/blob/47b3712898539569c02ec5b3ed4a6c36811331a1/src/peft/utils/integrations.py#L41
def dequantize_bnb_weight(weight: "torch.nn.Parameter", state=None):
    """
    Helper function to dequantize 4bit or 8bit bnb weights.

    If the weight is not a bnb quantized weight, it will be returned as is.
    """
    if not isinstance(weight, torch.nn.Parameter):
        raise TypeError(f"Input weight should be of type nn.Parameter, got {type(weight)} instead")

    cls_name = weight.__class__.__name__
    if cls_name not in ("Params4bit", "Int8Params"):
        return weight

    if cls_name == "Params4bit":
        output_tensor = bnb.functional.dequantize_4bit(weight.data, weight.quant_state)
        return output_tensor

    if state.SCB is None:
        state.SCB = weight.SCB

    if hasattr(bnb.functional, "int8_vectorwise_dequant"):
        # Use bitsandbytes API if available (requires v0.45.0+)
        dequantized = bnb.functional.int8_vectorwise_dequant(weight.data, state.SCB)
    else:
        # Multiply by (scale/127) to dequantize.
        dequantized = weight.data * state.SCB.view(-1, 1) * 7.874015718698502e-3

View on GitHub (pinned to a597f97485)

Solutions

1. Pass an `nn.Parameter` weight to the bitsandbytes layer replacement.
2. Wrap plain tensors with `nn.Parameter(tensor)`.

When it happens

Trigger: Raised in bitsandbytes quantization when the weight passed to a quantized linear is not an nn.Parameter.

Common situations: Passing plain tensors to bnb 8-bit/4-bit linear initialization instead of nn.Parameter.

AI-assisted analysis of huggingface/transformers@a597f97485 (2026-08-14). Data as JSON: /api/errors/8273724bf55a47d3. Report an issue: GitHub.