xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-14 05:35:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] Fix TP > 1 for new granite (#8544)

Browse Source 
Signed-off-by: Joe Runde <Joseph.Runde@ibm.com>
This commit is contained in:
Joe Runde 2024-09-17 17:17:08 -06:00 committed by GitHub
parent 56c3de018c
commit 98f9713399
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 2 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
vllm/model_executor/models/granite.py
View File

@ -428,6 +428,7 @@ class GraniteForCausalLM(nn.Module, SupportsLoRA):
sampling_metadata: SamplingMetadata) -> Optional[torch.Tensor]:
logits = self.logits_processor(self.lm_head, hidden_states,
sampling_metadata)
if logits is not None:
logits /= self.config.logits_scaling
return logits