Skip to content

fix return_token_log_probs on vLLM > 0.3.3 endpoints#498

Merged
saiatmakuri merged 3 commits intomainfrom
saiatmakuri/fix-return_token_log_probs
Apr 23, 2024
Merged

fix return_token_log_probs on vLLM > 0.3.3 endpoints#498
saiatmakuri merged 3 commits intomainfrom
saiatmakuri/fix-return_token_log_probs

Conversation

@saiatmakuri
Copy link
Copy Markdown
Contributor

Pull Request Summary

since vLLM > 0.3.3, logprobs are returned as as type Logprob instead as direct float's. it's an API-breaking change (see vllm-project/vllm#3065 (comment))

change vllm_server to return output in the same previous format

Test Plan and Usage Guide

create new llama-2-7b with vLLM==0.4.0.post1

Completion.create(
    model="llama-2-7b",
    prompt="Hello ",
    max_new_tokens=10,
    temperature=0.0,
    return_token_log_probs=True,
)

@saiatmakuri saiatmakuri requested a review from dmchoiboi April 23, 2024 04:30
Copy link
Copy Markdown
Contributor

@yunfeng-scale yunfeng-scale left a comment

Choose a reason for hiding this comment

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

thanks for the fix! i think we need to do some backfills also?

@saiatmakuri
Copy link
Copy Markdown
Contributor Author

thanks for the fix! i think we need to do some backfills also?

will update endpoints to latest image post merge

@saiatmakuri saiatmakuri merged commit 10d84ca into main Apr 23, 2024
@saiatmakuri saiatmakuri deleted the saiatmakuri/fix-return_token_log_probs branch April 23, 2024 05:05
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.

2 participants