Remove causal masks cache #412
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Code Metrics Report=============================================================================== Language Files Lines Code Comments Blanks =============================================================================== Dockerfile 1 34 25 0 9 Happy 1 442 369 0 73 JSON 9 21 21 0 0 Python 27 995 848 29 118 TOML 16 430 390 1 39 ------------------------------------------------------------------------------- Jupyter Notebooks 1 0 0 0 0 |- Markdown 1 60 30 22 8 |- Python 1 96 87 1 8 (Total) 156 117 23 16 ------------------------------------------------------------------------------- Markdown 16 1091 0 809 282 |- BASH 5 100 97 0 3 |- Python 6 122 110 0 12 |- Rust 2 80 72 3 5 (Total) 1393 279 812 302 ------------------------------------------------------------------------------- Rust 109 33248 30125 552 2571 |- Markdown 55 627 13 581 33 (Total) 33875 30138 1133 2604 =============================================================================== Total 181 36261 31778 1391 3092 =============================================================================== |
|
Looks good to me, I cannot reproduce the problem on this PR. For context - I observed that memory was not being fully freed in inference, so a long request might take up a bunch of memory and not properly release it, even when shorter requests come through. This fixes that -- thanks @EricLBuehler! |
|
@gregszumel thank you for confirming! |
EricLBuehler
added a commit
that referenced
this pull request
Jun 10, 2024
EricLBuehler
added a commit
that referenced
this pull request
Jun 10, 2024
* Implement gpt2 gguf tokenizer * Fix unk tok calculation * Remove normalizer * Update gguf tokenizer * Allow adding unk token when found * Add unk token to builder if provided. * Improve add_special_tokens * Use tokenizerx builder * Add useful comment Co-authored-by: Brennan Kinney <5098581+polarathene@users.noreply.github.com> * Bump version to 0.1.16 (#404) * Bump version to 0.1.17 * Fix version bump * Add and update template READMEs (#405) * Add readmes * Fix typos * Improve Rust docs (#406) * Expose phi3v loader and remove unused deps (#408) * Support format for mixtral where experts are in one tensor (#355) * Normal loading metadata for vision models (#409) * Phi 3 vision ISQ support (#410) * ISQ support for phi3v * Document it * Remove causal masks cache (#412) * Fix: use new slice_assign (#415) * Use new slice_assign * Fix dead image links * Fix Phi-3 GGUF (#414) * Fix kv head usage * Fix rope weights * Clippy * Work on the gpt2 conversion * Add comment * Add some tests * Update readme --------- Co-authored-by: Brennan Kinney <5098581+polarathene@users.noreply.github.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.