[XPU] add xpu branch in compressed_tensors_moe_w4a4_mxfp4 (#44540)

Signed-off-by: Zhu, Zufang <zufang.zhu@intel.com>
Signed-off-by: Kunshang Ji <jikunshang95@gmail.com>
Co-authored-by: Kunshang Ji <jikunshang95@gmail.com>
This commit is contained in:
zofia
2026-06-07 12:27:34 +08:00
committed by GitHub
co-authored by Kunshang Ji
parent 3bb46975bd
commit 6181e80fe0
2 changed files with 12 additions and 1 deletions
+2 -1
View File
@@ -41,7 +41,8 @@ steps:
python3 examples/basic/offline_inference/generate.py --model ibm-research/PowerMoE-3b --block-size 64 --enforce-eager -tp 2 &&
python3 examples/basic/offline_inference/generate.py --model ibm-research/PowerMoE-3b --block-size 64 --enforce-eager -tp 2 --enable-expert-parallel &&
python3 examples/basic/offline_inference/generate.py --model superjob/Qwen3-4B-Instruct-2507-GPTQ-Int4 --max-model-len 8192 &&
VLLM_XPU_FUSED_MOE_USE_REF=1 python3 examples/basic/offline_inference/generate.py --model Qwen/Qwen3-30B-A3B-Instruct-2507-FP8 --enforce-eager -tp 2
VLLM_XPU_FUSED_MOE_USE_REF=1 python3 examples/basic/offline_inference/generate.py --model Qwen/Qwen3-30B-A3B-Instruct-2507-FP8 --enforce-eager -tp 2 --max-model-len 8192 &&
python3 examples/basic/offline_inference/generate.py --model INCModel/Qwen3-30B-A3B-Instruct-2507-MXFP4-LLMC --enforce-eager -tp 2 --max-model-len 8192
'
- label: "XPU V1 test"
depends_on:
@@ -21,6 +21,9 @@ from vllm.model_executor.layers.fused_moe.experts.cutlass_moe import (
from vllm.model_executor.layers.fused_moe.experts.marlin_moe import (
MarlinExperts,
)
from vllm.model_executor.layers.fused_moe.experts.xpu_moe import (
XPUExpertsMXFp4,
)
from vllm.model_executor.layers.fused_moe.oracle.mxfp4 import (
Mxfp4MoeBackend,
make_mxfp4_moe_kernel,
@@ -33,6 +36,7 @@ from vllm.model_executor.layers.quantization.utils.marlin_utils_fp4 import (
prepare_moe_fp4_layer_for_marlin,
)
from vllm.model_executor.utils import set_weight_attrs
from vllm.platforms import current_platform
logger = init_logger(__name__)
@@ -48,6 +52,10 @@ class CompressedTensorsW4A4Mxfp4MoEMethod(CompressedTensorsMoEMethod):
if self.use_cutlass_mxfp4:
logger.info_once("Using CutlassExpertsMxfp4 for MXFP4 MoE")
self.experts_cls = CutlassExpertsMxfp4
elif current_platform.is_xpu():
self.mxfp4_backend = Mxfp4MoeBackend.XPU
self.experts_cls = XPUExpertsMXFp4
logger.info_once("Using XPUExpertsMXFp4 for MXFP4 MoE on XPU platform")
else:
logger.info_once("Using MarlinExperts for MXFP4 MoE")
self.experts_cls = MarlinExperts
@@ -179,6 +187,8 @@ class CompressedTensorsW4A4Mxfp4MoEMethod(CompressedTensorsMoEMethod):
layer.w2_weight_scale = torch.nn.Parameter(
torch.stack(swizzled_w2), requires_grad=False
)
elif current_platform.is_xpu():
pass
else:
logger.warning_once(
"Your GPU does not have native support for FP4 computation "