[Quantization] Select linear backends per quantization (#51204)

Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
This commit is contained in:
Netanel Haber
2026-09-14 16:24:18 -04:00
committed by GitHub
parent 995e8581f4
commit d4ee7fe7a9
6 changed files with 154 additions and 17 deletions
@@ -158,7 +158,7 @@ class AiterMxfp4LinearKernel(MxFp4LinearKernel):
from vllm._aiter_ops import is_aiter_found_and_supported
from vllm.model_executor.kernels.linear import _get_linear_backend
linear_backend = _get_linear_backend()
linear_backend = _get_linear_backend(quantization="mxfp4")
if (
current_platform.is_rocm()