mirror of
https://git.datalinker.icu/vllm-project/vllm.git
synced 2026-03-23 01:59:11 +08:00
b6520e5ec6
Fixes #29525 RTN (Round-To-Nearest) quantization is currently not supported on ROCm. This fix adds a check to skip RTN quantization tests when running on ROCm to prevent CI failures for unsupported features. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com> Signed-off-by: westers <steve.westerhouse@origami-analytics.com>
23 lines
814 B
Python
23 lines
814 B
Python
# SPDX-License-Identifier: Apache-2.0
|
|
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
|
|
|
from vllm.model_executor.layers.quantization import get_quantization_config
|
|
from vllm.platforms import current_platform
|
|
|
|
|
|
def is_quant_method_supported(quant_method: str) -> bool:
|
|
# Currently, all quantization methods require Nvidia or AMD GPUs
|
|
if not (current_platform.is_cuda() or current_platform.is_rocm()):
|
|
return False
|
|
|
|
# RTN quantization is currently not supported on ROCm
|
|
if current_platform.is_rocm() and quant_method == "rtn":
|
|
return False
|
|
|
|
capability = current_platform.get_device_capability()
|
|
assert capability is not None
|
|
|
|
min_capability = get_quantization_config(quant_method).get_min_capability()
|
|
|
|
return capability.to_int() >= min_capability
|