xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-03-20 06:41:19 +08:00
Code Issues Packages Projects Releases Wiki Activity

[ROCM] Enable CompressedTensorsWNA16 (#27187)

Browse Source 
Signed-off-by: JartX <sagformas@epdcenter.es>
This commit is contained in:
JartX 2025-10-21 16:43:23 +02:00 committed by GitHub
parent bd66b8529b
commit ba09652de2
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 4 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors_moe.py
View File

@ -142,7 +142,10 @@ class CompressedTensorsMoEMethod(FusedMoEMethodBase):
# group_size=None means channelwise
group_size = weight_quant.group_size or -1
# Prefer to use the MarlinMoE kernel when it is supported.
if not check_moe_marlin_supports_layer(layer, group_size):
if (
not check_moe_marlin_supports_layer(layer, group_size)
or current_platform.is_rocm()
):
if (
weight_quant.strategy == QuantizationStrategy.GROUP
and weight_quant.actorder