wassname/vllm
1
0 Fork 0
mirror of https://github.com/wassname/vllm.git synced 2026-10-29 13:50:10 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Speculative decoding] [Bugfix] Fix overallocation in ngram + spec logprobs (#4672)

Browse Source
This commit is contained in:
Cade Daniel authored and GitHub committed 2024-05-08 23:24:46 +00:00
1 parent f942efb5a3
commit 8b9241be3a
1 file changed
+1 -1
Download Patch File Download Diff File Expand all files Collapse all files
vllm/spec_decode/ngram_worker.py
+1 -1
View File
@@ -138,7 +138,7 @@ class NGramWorker(LoraNotSupportedWorkerBase):
SamplerOutput(
outputs=None,
sampled_token_probs=token_probs[i],
logprobs=token_logprobs,
logprobs=token_logprobs[i],
sampled_token_ids=token_ids[i],
))
return outputs, False