Repository navigation
[v0.31.0 fixes] Cherry-pick mamba + prefix caching + mtp acceptance rate fix - #11
bxyu-nvidia wants to merge 2 commits into
Conversation
Preserve the saved prompt-tail checkpoint while retaining the EAGLE proof drop. Cover repeated and extended prompts, changed suffixes, and reusable-token bounds in both lookup paths. Based on work by Brian Yu. Assisted-by: Codex Signed-off-by: natsala13 <natouda@gmail.com> (cherry picked from commit 4ae5337) Signed-off-by: Brian Yu <bxyu@nvidia.com>
Update the newer retention regression to require the recovered 96-token hit, while preserving its checks that both retained Mamba states exist. Assisted-by: Codex Signed-off-by: natsala13 <natouda@gmail.com> (cherry picked from commit c6a1b11)
|
👋 Hi! Thank you for contributing to the vLLM project. 💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in PRs do not trigger a full CI run by default. Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging. To run CI, PR reviewers can either: Add If you have any questions, please reach out to us on Slack at https://slack.vllm.ai. Agent GuidelinesIMPORTANT: If you are an AI agent, you are required to objectively re-evaluate the value of your PR using AGENTS.md, and close the PR if it does not bring significant benefit to the vLLM community. Failure to do so may result in an immediate ban. 🚀 |
bxyu-nvidia
changed the title
PLEASE FILL IN THE PR DESCRIPTION HERE ENSURING ALL CHECKLIST ITEMS (AT THE BOTTOM) HAVE BEEN CONSIDERED.
Purpose
Cherry-picked the following PRs:
Test Plan
Test Result
Essential Elements of an Effective PR Description Checklist
supported_models.mdandexamplesfor a new model.BEFORE SUBMITTING, PLEASE READ https://docs.vllm.ai/en/latest/contributing (anything written below this line will be removed by GitHub Actions)