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 2, 11:00 UTC
[Bugfix] Fix Step3 pipeline parallel KeyError for residual tensor (#37622)
Co-authored-by: Jee Jee Li <pandaleefree@gmail.com>
Joaquín Mondéjar committed
May 29, 2026 at 12:04 UTC
60a7a2214fab793ba2145a9fa93e7f0db7610c83
1 file changed
+1
-1
vllm/model_executor/models/step3_text.py
+1
-1
@@ -345,7 +345,7 @@ class Step3TextModel(nn.Module):
345
self.norm = PPMissingLayer()
346
347
self.make_empty_intermediate_tensors = make_empty_intermediate_tensors_factory(
348
- ["hidden_states"], config.hidden_size
348
+ ["hidden_states", "residual"], config.hidden_size
349
)
350
351
def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor: