Artifact-evaluation fixes: is_moe, judge comments, LICENSE, GPU params, manifest, checksums, appendix PDFs - #12
Merged
Conversation
… checksums, fewer-GPU docs - Launchers: make GPU count overridable via $NPROC_PER_NODE and $CUDA_VISIBLE_DEVICES instead of hard-coded --nproc_per_node / export lines; defaults preserved (Review D). - datasets/MANIFEST.md: manifest of every dataset each training/baseline/eval script consumes, with source (Zenodo) and how to regenerate (Review D). - CHECKPOINTS.md: pinned-revision + SHA-256 checksum manifest and verify script for the released LoRA adapters (Review D). - README: document NPROC_PER_NODE/CUDA_VISIBLE_DEVICES overrides and fewer/smaller-GPU training (batch/grad-accum, single-GPU, 24GB QLoRA) (Review B); link manifest+checksums.
… obtain checksums/revisions
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Carries the CCS AEC review fixes and packaging that landed after PR #10 was merged (PR #10 merged the earlier
cb73b5asnapshot).Review C
data_generation/data_curation_drip.py) and PISmith reward auditor (testing/pismith/utils.py,train_sep.py) check data inclusion / witness presence, not whether the injected task was executed. Fixed the misleading "executed" wording in code +data_generation/pismithREADMEs. (SEP-eval judge left as-is — it correctly checks execution.)Review D
is_moeflag that broke DRIP training (train_unified.py); keep the dense LoRA path, drop the unused MoE branch.LICENSEclaimed in the appendix.$NPROC_PER_NODE/$CUDA_VISIBLE_DEVICES(defaults preserved).datasets/MANIFEST.md: every dataset each training/baseline/eval script consumes, its source (Zenodo), and how to regenerate.CHECKPOINTS.md: pinned revisions + SHA-256 checksums for the four released LoRA adapters, with a verify script.Review B
Author-requested
Artifact_Appendix.pdfandDRIP_Supplementary.pdfwith README references.🤖 Generated with Claude Code
Generated by Claude Code