[Model] Pass intermediate_tensors to the model when capturing CUDA gr… (#57745)
Co-authored-by: jiazhimi <jiazhimi@tencent.com>
Mi Jiazhi committed
Sep 20, 2026 at 14:23 UTC
e378275a8f20eac9e92212e4a1fee84ec5a31dc7
1 file changed
+1
vllm/v1/worker/gpu/cudagraph_utils.py
+1
@@ -662,6 +662,7 @@ class ModelCudaGraphManager(CudaGraphManager):
662
model_inputs = {
663
"input_ids": input_buffers.input_ids[:num_tokens],
664
"positions": input_buffers.positions[:num_tokens],
665
+ "intermediate_tensors": None,
666
**model_state.prepare_dummy_inputs(num_reqs, num_tokens),
667
}
668
if not self.is_first_pp_rank: