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 16, 04:30 UTC
[Bugfix]: Clamp `-inf` logprob values in prompt_logprobs (#11073)
Signed-off-by: Rafael Vasquez <rafvasq21@gmail.com>
Rafael Vasquez committed
Dec 11, 2024 at 04:27 UTC
40766ca1b8b0ef92e220595bda96c4336b597e5b
1 file changed
+6
vllm/entrypoints/openai/serving_completion.py
+6
@@ -392,6 +392,12 @@ class OpenAIServingCompletion(OpenAIServing):
392
prompt_token_ids = final_res.prompt_token_ids
393
assert prompt_token_ids is not None
394
prompt_logprobs = final_res.prompt_logprobs
395
+ if prompt_logprobs:
396
+ for logprob_dict in prompt_logprobs:
397
+ if logprob_dict:
398
+ for logprob_values in logprob_dict.values():
399
+ if logprob_values.logprob == float('-inf'):
400
+ logprob_values.logprob = -9999.0
401
prompt_text = final_res.prompt
402
403
token_ids: GenericSequence[int]