Skip to content

Support chunked prefill with prefix KV cache reuse in Tunix Gemma 4. - #1906

Open
copybara-service[bot] wants to merge 1 commit into
mainfrom
test_963446157
Open

Support chunked prefill with prefix KV cache reuse in Tunix Gemma 4.#1906
copybara-service[bot] wants to merge 1 commit into
mainfrom
test_963446157

Conversation

@copybara-service

Copy link
Copy Markdown

Support chunked prefill with prefix KV cache reuse in Tunix Gemma 4.

  • Add prefix Key/Value cache reuse for chunked prefill across global linear caches and sliding-window ring buffers.
  • Implement prefix length bucketing to prevent XLA recompilation storms.
  • Advance cache end index strictly by non-PAD token counts with ragged mask support.

@google-cla

google-cla Bot commented Aug 17, 2026

Copy link
Copy Markdown

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

- Add prefix Key/Value cache reuse for chunked prefill across global linear caches and sliding-window ring buffers.
- Implement prefix length bucketing to prevent XLA recompilation storms.
- Advance cache end index strictly by non-PAD token counts with ragged mask support.

PiperOrigin-RevId: 963446157
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.

1 participant