Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

perplexity : fix kv cache handling for hellaswag #4981

Merged
merged 1 commit into from
Jan 16, 2024
Merged

perplexity : fix kv cache handling for hellaswag #4981

merged 1 commit into from
Jan 16, 2024

Conversation

ggerganov
Copy link
Owner

ref #2321

Clear KV cache from previous chunk when running Hellaswag

@JianbangZ
Copy link

Working now. Hellaswag score for MIstral-7B-v0.1 is 81, same as what's officially reported.

ikawrakow
ikawrakow approved these changes Jan 16, 2024
View reviewed changes
Copy link
Contributor

@ikawrakow ikawrakow left a comment

Choose a reason for hiding this comment

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

This works, but the result is very slightly different from running #4982. Why?

@ikawrakow ikawrakow mentioned this pull request Jan 16, 2024
Closed
@ggerganov
Copy link
Owner Author

Not 100% sure. In general, evaluating a prompt "[a][b]" where [a] and [b] are some arrays of tokens, will produce slightly different results compared to first evaluating "[a]" and then "[b]", but I'm not sure if it's enough to explain the variation that you observe

@ggerganov ggerganov merged commit 959ef0c into master Jan 16, 2024
53 checks passed
@ggerganov ggerganov deleted the gg/hellaswag-clear-kv-cache branch January 16, 2024 17:34
@MarcusDunn MarcusDunn mentioned this pull request Jan 16, 2024
Closed
jordankanter pushed a commit to jordankanter/llama.cpp that referenced this pull request Feb 3, 2024
@ggerganov @jordankanter
perplexity : fix kv cache handling for hellaswag (ggerganov#4981)
4562e5f 
ggml-ci
hodlen pushed a commit to hodlen/llama.cpp that referenced this pull request Apr 1, 2024
@ggerganov @hodlen
perplexity : fix kv cache handling for hellaswag (ggerganov#4981)
1eabefe 
ggml-ci
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@ikawrakow ikawrakow ikawrakow approved these changes

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@ggerganov @JianbangZ @ikawrakow