Skip to content

Commit 7485451

Browse files
committed
Fix pre-commit (and XPU) on main
- vllm-project#28112 didn't cause conflicts but main still contained a `use_v1` - Also fixes a pre-commit warning Signed-off-by: Harry Mellor <[email protected]>
1 parent 10138c9 commit 7485451

File tree

2 files changed

+1
-3
lines changed

2 files changed

+1
-3
lines changed

vllm/model_executor/layers/quantization/mxfp4.py

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1145,7 +1145,7 @@ def apply(
11451145
) -> torch.Tensor:
11461146
assert activation == "swigluoai", (
11471147
"Only swiglu_oai activation is supported for IPEX MXFP4 MoE"
1148-
) # noqa:
1148+
)
11491149
hidden_size_pad = round_up(self.original_hidden_size, 128)
11501150
x_pad = torch.nn.functional.pad(x, (0, hidden_size_pad - x.size(-1)))
11511151
hidden_states = layer.ipex_fusion(

vllm/platforms/xpu.py

Lines changed: 0 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -64,8 +64,6 @@ def get_attn_backend_cls(
6464

6565
if use_sparse:
6666
raise NotImplementedError("Sparse Attention is not supported on XPU.")
67-
if not use_v1:
68-
raise ValueError("XPU backend only supports V1.")
6967
if selected_backend == AttentionBackendEnum.TRITON_ATTN:
7068
logger.info_once("Using Triton backend.")
7169
return AttentionBackendEnum.TRITON_ATTN.get_path()

0 commit comments

Comments
 (0)