Skip to content

Conversation

@realliujiaxu
Copy link
Collaborator

What this PR does / why we need it?

fix greedy temperature detection from vllm-project/vllm#27077

Does this PR introduce any user-facing change?

No

How was this patch tested?

Signed-off-by: realliujiaxu <[email protected]>
Copy link
Contributor

@gemini-code-assist gemini-code-assist bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Code Review

This pull request addresses a bug in greedy sampling detection by changing the GREEDY_TEMPERATURE constant from -1 to 0. A temperature of 0 is the standard for greedy decoding, ensuring the model selects the highest probability token. This change aligns the implementation with common practice and the project's own unit tests, and is a correct and necessary fix for speculative decoding functionality.

@realliujiaxu realliujiaxu marked this pull request as draft December 27, 2025 04:51
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant