Repository navigation
fix: count global layers in hybrid KV-cache estimates - #18
Merged
Merged
Conversation
Contributor
Author
|
Hi, could you please approve the CI and Action self-test workflows when you have a moment? Thanks for taking a look! |
This file contains hidden or 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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Hybrid KV-cache estimates now count global layers at the full context length. Explicit layer types take precedence over the sliding-window pattern, while the existing Gemma 2 fallback and fully sliding models keep their behavior.
Added regression coverage for GPT-OSS, Gemma 3 and malformed configs. Locally, 7549 tests passed with one Windows symlink skip. Profile validation, the benchmark gate and the site build passed. The benchmark's three declared gaps are unchanged from the base branch.
Model references: GPT-OSS configuration and Gemma 3 layer pattern.
Fixes #7