Repository navigation
Add chat RULER+ at 64K to the post-training dev suites - #561
Open
abhishekraok wants to merge 2 commits into
Open
abhishekraok wants to merge 2 commits into
abhishekraok wants to merge 2 commits into
Conversation
RULER+ gets a :chat_mt512 variant (one chat turn, 512 output tokens, items, greedy
decoding, stops and scoring unchanged) and a ruler_plus_all__{size}:hillclimb:chat_mt512
suite. Each dev size gets posttrain:{size}:dev:chat_ruler, run as its own job with a
think-off chat template.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Codex Review Summary: ✅ CompletedThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
This was referenced Nov 12, 2026
# Conflicts: # tests/evals/suites/test_posttrain.py
This branch has not been deployed
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Written by Claude Code session 27c852a2-737f-40d7-a147-784d52aa3a9b
Adds a chat-format RULER+ read to the post-training dev suites: a follow-up to posttrain-suites (#552).
Why. Completion RULER+ understates what post-training costs in the format the model is served in. The post-training ledger measured this on the original RULER at 64K, on Olmo 3.5 SFT checkpoints, with thinking off and 512 output tokens. On the same items, chat scored 4–6 points below completion on the 7 tasks whose answers fit the cap, and 8–12 points below on all 13.
What.
ruler_plus_*:chat_mt512sends the completion input as one user turn, asruler_*:chatdoes, with 512 output tokens. Items, greedy decoding, stops and scoring are unchanged, so it pairs item-for-item with the completion cell.ruler_plus_all__{size}:hillclimb:chat_mt512suites, on 128 items per task.posttrain:{tiny,small}:dev:chat_rulerholds the 64K suite. Run it as its own job, serving a think-off chat template (-o provider.tokenizer=<dir>). With thinking on, 512 tokens cut off the reasoning, and even at 8,192 tokens 70% of answers hit the cap.Checks.
🤖 Generated with Claude Code