Send routed_experts_start_len per turn; merge per-turn deltas positionally - #82
Draft
nightlessbaron wants to merge 36 commits into
Draft
Send routed_experts_start_len per turn; merge per-turn deltas positionally#82nightlessbaron wants to merge 36 commits into
nightlessbaron wants to merge 36 commits into
Conversation
… samples Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…fetch Re-tokenizing the full trajectory on the session-server event loop blocks record retrieval at long context; gate it behind --tito-session-mismatch-sample-rate (default off). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> (cherry picked from commit 759ca30)
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…removed-zero-mask
…nto feat/validator-allow-removed-zero-mask # Conflicts: # miles/rollout/session/sessions.py
…nto feat/validator-allow-removed-zero-mask
…nto feat/validator-allow-removed-zero-mask
…nto feat/validator-allow-removed-zero-mask
(cherry picked from commit dba65d3)
This reverts commit cf6de42.
…er to match previous implementation
nightlessbaron
force-pushed
the
feat/routed-experts-delta
branch
from
July 26, 2026 09:52
3dab1bb to
14a7daf
Compare
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Companion to LLM360/sglang#45. For
return_routed_expertsrequests the session server now sendsrouted_experts_start_len = len(checkpoint) - max_trim_tokens - 1, so the engine returns only the expert rows for newly processed tokens. The merge assembles deltas positionally: truncate the buffer at the record's offset, append its rows (offset 0 = legacy full-echo last-wins, so the path is backward compatible). Record compaction preserves the offset field, and the existing full-coverage assert still guards the final tensor.Effect on rollout throughput (per-task LLM vs env time, before/after):
Note for review: the branch sits on the session-server-side merge stack (earlier commits); the delta change itself is the tip commit.