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 11, 08:45 UTC
[Bugfix][CPU] Fix CPU backend ROPE dispatch for VL models (#30829)
Signed-off-by: jiang1.li <jiang1.li@intel.com> Signed-off-by: Li, Jiang <bigpyj64@gmail.com> Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Li, Jiang committed
Dec 17, 2025 at 15:25 UTC
0cd5353644d3d045ab33c7e8e19c182bfd7db911
1 file changed
+9
vllm/model_executor/layers/rotary_embedding/common.py
+9
@@ -264,6 +264,15 @@ class ApplyRotaryEmb(CustomOp):
264
265
return output
266
267
+ def forward_cpu(
268
+ self,
269
+ x: torch.Tensor,
270
+ cos: torch.Tensor,
271
+ sin: torch.Tensor,
272
+ ) -> torch.Tensor:
273
+ # TODO (bigPYJ1151): need to enable fused CPU ROPE here
274
+ return self.forward_native(x, cos, sin)
275
+
276
def extra_repr(self) -> str:
277
s = f"is_neox_style={self.is_neox_style}"
278
s += f"enable_fp32_compute={self.enable_fp32_compute}"