biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Core] Switch Flat logprob control from environment variable to SamplingParams (#28914)

Browse Source 
Signed-off-by: Jialin Ouyang <Jialin.Ouyang@gmail.com>
Co-authored-by: 22quinn <33176974+22quinn@users.noreply.github.com>
This commit is contained in:
Jialin Ouyang
2025-11-18 18:10:02 -08:00
committed by GitHub
parent da94c7c0eb
commit 40b6b38f2c
6 changed files with 33 additions and 41 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
vllm/logprobs.py
View File

@@ -5,8 +5,6 @@ from collections.abc import Iterable, Iterator, MutableSequence
from dataclasses import dataclass, field
from typing import overload
import vllm.envs as envs
# We use dataclass for now because it is used for
# openai server output, and msgspec is not serializable.
@@ -161,17 +159,17 @@ PromptLogprobs = FlatLogprobs | list[LogprobsOnePosition | None]
SampleLogprobs = FlatLogprobs | list[LogprobsOnePosition]
def create_prompt_logprobs() -> PromptLogprobs:
def create_prompt_logprobs(flat_logprobs: bool) -> PromptLogprobs:
"""Creates a container to store prompt logprobs for a request"""
logprobs = FlatLogprobs() if envs.VLLM_FLAT_LOGPROBS else []
logprobs = FlatLogprobs() if flat_logprobs else []
# NOTE: logprob of first prompt token is None.
logprobs.append(None)
return logprobs
def create_sample_logprobs() -> SampleLogprobs:
def create_sample_logprobs(flat_logprobs: bool) -> SampleLogprobs:
"""Creates a container to store decode logprobs for a request"""
return FlatLogprobs() if envs.VLLM_FLAT_LOGPROBS else []
return FlatLogprobs() if flat_logprobs else []
def append_logprobs_for_next_position(