[Bugfix][MTP] Integrate safe YaRN context inheritance (#410) - #540
Merged
yangzhuxinyzx merged 6 commits intoSep 7, 2026
Merged
Conversation
Assisted-by: OpenAI Codex
…audit-20260907-031538 Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
Integrate #410 on current main. Keep malformed test inputs out of Transformers constructor validation and test finite limits, target ceilings and unrelated-drafter isolation. Co-authored-by: OpenAI Codex <noreply@openai.com> Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
…yarn-audit-20260907-031538 Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
Only inherit target YaRN when existing draft scaling does not already cover the request; retain valid linear and checkpoint YaRN configurations. Co-authored-by: OpenAI Codex <noreply@openai.com> Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
…yarn-audit-20260907-031538 Signed-off-by: yangzhuxinyzx <153831768+yangzhuxinyzx@users.noreply.github.com>
yangzhuxinyzx
marked this pull request as ready for review
September 7, 2026 03:29
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Purpose
Integrate #410 against latest public main
586026e7c575febd6f2f321bf7df17bbbb7e98fe, preserving its original commits. Inherit validated explicit target YaRN for an extended native MTP drafter, and eliminate changing-argument warning spam above its configured cap.Audit repairs
Test Plan / Result
19 targeted CPU/meta configuration tests passed, including the real ModelConfig length derivation helper for existing scaling. Scoped pre-commit including mypy passed. No model download, server startup, full E2E or new GPU benchmark. Await fresh GitHub integration CI before ready/merge.
Boundaries
This is configuration and warning behavior, not a new attention kernel or new numerical approximation. The original PR's historical 564K MTP service evidence is retained but is not a fresh latest-main benchmark or proof of 1M output quality. Only explicitly requested extended native MTP configuration inherits new scaling; unrelated drafts and unextended contexts remain unchanged.
Ownership / evidence
Owned worktree
/home/ymzx/桌面/1cat-vllm/worktrees/v100-sep07-mtp-yarn-audit-20260907-031538. Logs/tmp/1cat-sep07-yarn-cpu.log(original failures),/tmp/1cat-sep07-yarn-cpu4.log(19 passed),/tmp/1cat-sep07-yarn-commit4.log(static gates). Canonical dirty checkout untouched.