[Online quantization] Add online MXFP4 quantization support (#49347)

Signed-off-by: Felix Marty <Felix.Marty@amd.com>
This commit is contained in:
fxmarty-amd
2026-08-07 12:39:46 -07:00
committed by GitHub
parent a671679e9f
commit 3518110f2f
18 changed files with 1423 additions and 35 deletions
@@ -23,6 +23,7 @@ from vllm.model_executor.layers.quantization.utils.quant_utils import (
kMxfp6E3M2Dynamic,
)
from vllm.platforms import current_platform
from vllm.utils.import_utils import has_quark
from .base import MxFp4LinearKernel, MxFp4LinearLayerConfig
@@ -68,6 +69,9 @@ class EmulationMxfp4LinearKernel(MxFp4LinearKernel):
):
return False, "only supports MXFP4 or MXFP6 or unquantized activations"
if not has_quark():
return False, "amd-quark package not available"
if (
current_platform.is_rocm()
and current_platform.supports_mx()