vllm-project/vllm · error · ValueError

The quantization method %s is deprecated and will be removed

Error message

The quantization method %s is deprecated and will be removed in future versions of vLLM. To bypass, set `--allow-deprecated-quantization`.

What it means

Raised when the requested quantization method is listed in DEPRECATED_QUANTIZATION_METHODS and the user has not opted in with allow_deprecated_quantization. vLLM removes old quantization formats over time; this error forces users to acknowledge the pending removal. With the flag set, the same condition only logs a warning and startup continues.

Source

Thrown at vllm/config/model.py:1296

                )

        if self.quantization is not None:
            if self.quantization not in supported_quantization:
                raise ValueError(
                    f"Unknown quantization method: {self.quantization}. Must "
                    f"be one of {supported_quantization}."
                )
            current_platform.verify_quantization(self.quantization)

        if self.quantization in me_quant.DEPRECATED_QUANTIZATION_METHODS:
            if self.allow_deprecated_quantization:
                logger.warning(
                    "The quantization method %s is deprecated "
                    "and will be removed in future versions of vLLM.",
                    self.quantization,
                )
            else:
                raise ValueError(
                    "The quantization method %s is deprecated "
                    "and will be removed in future versions of vLLM. To bypass, "
                    "set `--allow-deprecated-quantization`.",
                    self.quantization,
                )

    def _verify_cuda_graph(self) -> None:
        # CUDAGraph capture not supported for encoder-decoder models on ROCm
        unsupported_rocm = self.is_encoder_decoder
        if unsupported_rocm and not self.enforce_eager and current_platform.is_rocm():
            logger.warning(
                "CUDA graph is not supported for %s on ROCm yet, fallback "
                "to eager mode.",
                self.model_arch_config.model_type,
            )
            self.enforce_eager = True

    def _verify_with_expert_parallelism(self) -> None:

View on GitHub (pinned to c794754062)

Solutions

Switch the checkpoint to a currently supported quantization method (re-quantize or download a newer checkpoint build).
If you accept the removal risk, launch with --allow-deprecated-quantization (CLI) or allow_deprecated_quantization=True (Llm/AsyncEngineArgs).
Check the vLLM release notes / changelog for the deprecation announcement and the recommended replacement format.

Example fix

# before
vllm serve my-model --quantization <deprecated-method>
# after
vllm serve my-model --quantization <deprecated-method> --allow-deprecated-quantization

Defensive patterns

Strategy: validation

Validate before calling

from vllm.model_executor.models import me_quant
def check_quant_not_deprecated(q: str) -> bool:
    return q not in me_quant.DEPRECATED_QUANTIZATION_METHODS
# before LLM(...): if deprecated, either re-quantize or pass
# allow_deprecated_quantization=True deliberately.

Try / catch

except ValueError as e:
    if 'deprecated' in str(e) and 'allow-deprecated-quantization' in str(e):
        retry with allow_deprecated_quantization=True after logging a removal-risk warning

Prevention

Track vLLM release notes for quantization deprecation announcements before upgrading.
Prefer re-quantizing checkpoints to supported formats instead of relying on the bypass flag.
Assert your quantization method against DEPRECATED_QUANTIZATION_METHODS in CI startup checks.

When it happens

Trigger: Launching a model whose --quantization (or HF config quantization_config) resolves to a method in vllm/model_executor/models/me_quant.DEPRECATED_QUANTIZATION_METHODS, without passing --allow-deprecated-quantization.

Common situations: Upgrading vLLM to a release that deprecates a quantization format (e.g. an old AWQ/GPTQ variant) while serving checkpoints quantized with it; loading a community checkpoint whose config.json embeds the deprecated method name.

Related errors

AI-assisted analysis of vllm-project/vllm@c794754062 (2026-08-14). Data as JSON: /api/errors/1c390a8d5c72371e. Report an issue: GitHub.