A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 3, 11:45 UTC
[XPU] fix weight scale shape (#42725)
Signed-off-by: Zhu, Zufang <zufang.zhu@intel.com> Co-authored-by: Kunshang Ji <kunshang.ji@intel.com>
zofia committed
May 17, 2026 at 16:55 UTC
0fa888465e5a30b797bdf2cdcd0f57fc77541cef
1 file changed
+3
vllm/model_executor/kernels/linear/scaled_mm/xpu.py
+3
@@ -59,6 +59,9 @@ class XPUFP8ScaledMMLinearKernel(FP8ScaledMMLinearKernel):
59
replace_parameter(layer, "weight", weight.data.t())
60
# else: already in [in, out] layout — no-op
61
62
+ weight_scale = layer.weight_scale.t().contiguous()
63
+ replace_parameter(layer, "weight_scale", weight_scale.data)
64
+
65
def apply_weights(
66
self,
67
layer: torch.nn.Module,