add ogbench submodule#11
Merged
Merged
Conversation
helenlu66
force-pushed
the
codex/add-ogbench-submodule
branch
from
May 15, 2026 16:38
31baac6 to
4260c12
Compare
Contributor
There was a problem hiding this comment.
Pull request overview
Adds the forked ogbench repository as a Git submodule so the main repo can reference it as an external dependency without vendoring its contents.
Changes:
- Introduces a new
.gitmodulesentry for theogbenchsubmodule. - Configures the submodule path (
ogbench), remote URL, and tracking branch (master).
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
seanrivera
added a commit
that referenced
this pull request
Jul 2, 2026
…conditional (exp-3) sweep Preflight-harden the 10-config conditional sweep + config #11 for a paid launch. Fixes three launch-fatal blockers found in preflight and two more caught live by the M6 API smoke. - Opus 4.8 as the Claude model: interface/agents/claude.py omits temperature for the Opus-4.7+/Fable/Mythos family (they 400 on sampling params) and gains opt-in adaptive thinking + output_config.effort. The 6 configs rename the model key claude_sonnet -> claude_opus, use claude-opus-4-8 at effort:low. - Launcher conditions plumbing: CONDITIONS/PROMPT_VARIANT env thread through lib/distributed_start.sh:start_coordinator into coordinator-prepare (guarded appends), so run_pipeline's H1/H2 guard no longer rejects a conditional prepare. - Kimi: the agent pins temperature by thinking mode (1.0 on / 0.6 off) — Moonshot 400s on any other value. The 10 configs run Kimi thinking-OFF (thinking-on spends the whole budget on reasoning and truncates before FINAL_OUTPUT at 4096, found via the M6 smoke); config #11 keeps Kimi thinking-on at 16384. - Qwen -> Qwen3.6-27B FP16 via vLLM on the qwen-fp16-80 image; FP8/3.5-HF retired from the conditional configs, the smoke_qwen36 config, and the deploy default. - Add config #11 (baseline re-run with full thinking: Opus xhigh, paid 8192/16384) to isolate reasoning-depth value vs the effort:low sweep. Pin DIFFICULTY_MAX=1000. M6 API smoke (V01 empty_room, live) validated: Opus 4.8 solves near-optimally with no 400s; Kimi thinking-off parses 33/33 turns clean. Full suite green (697). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MskZpoBtRRi93NFQBJ3xBN
seanrivera
added a commit
that referenced
this pull request
Jul 2, 2026
The 11 unique configs were spread across the 16-slot variant inventory and the 9-batch dedup table (batch 1 alone yields 3 configs), so no single place listed them. Add a numbered 1-11 table with the per-model thinking profile, making the 10 effort:low configs vs config #11 (the thinking probe) obvious at a glance. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MskZpoBtRRi93NFQBJ3xBN
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.
This is simply adding the ogbench forked repo as a git submodule. There are no commits inside the ogbench repo yet.