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.

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

Feature: Added api for getting/setting the kv_cache #685

Merged
merged 6 commits into from
Apr 2, 2023

Conversation

chrfalch
Copy link
Contributor

@chrfalch chrfalch commented Apr 1, 2023

This PR contains a simple extension to the C-API for getting/setting the kv_cache so that an app can save the state of the kv_cache after providing a prompt and load this next time the app starts to avoid having to evaluate the prompt on startup.

The api provides access methods for retrieving the current memory buffer for the kv_cache and its token number. It also contains a method for setting the kv_cache from a memory buffer and a token count.

I did not start implementing the --cache-prompt argument since it is a bit more involved - we need to save some more state like the last_n_tokens and n_past parameter. We'd also need to hash the prompt, check if a prompt file existed etc.

Implements foundation for #64

The api provides access methods for retrieving the current memory buffer for the kv_cache and its token number.
It also contains a method for setting the kv_cache from a memory buffer.

This makes it possible to load/save history - maybe support --cache-prompt paramater as well?
@chrfalch
Copy link
Contributor Author

chrfalch commented Apr 1, 2023

Example app using the new api to load the state is able to answer with previous knowledge from the saved prompt (uses the chat example):

image

Copy link
Collaborator

@prusnak prusnak left a comment

Choose a reason for hiding this comment

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

use const where appropriate

llama.cpp Outdated Show resolved Hide resolved
llama.cpp Outdated Show resolved Hide resolved
llama.h Outdated Show resolved Hide resolved
llama.h Outdated Show resolved Hide resolved
chrfalch and others added 4 commits April 1, 2023 18:46
Add review comments

Co-authored-by: Pavol Rusnak <pavol@rusnak.io>
Added review comments

Co-authored-by: Pavol Rusnak <pavol@rusnak.io>
Added review comments

Co-authored-by: Pavol Rusnak <pavol@rusnak.io>
Review Comments

Co-authored-by: Pavol Rusnak <pavol@rusnak.io>
@chrfalch
Copy link
Contributor Author

chrfalch commented Apr 1, 2023

Thanks for the comments, @prusnak - appreciate it.

@chrfalch chrfalch requested a review from prusnak April 1, 2023 17:30
Copy link
Owner

@ggerganov ggerganov left a comment

Choose a reason for hiding this comment

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

Very good start for solving #64

Use 4 spaces indentation

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.

3 participants