Repository navigation
Add bounded GRPO constructor diagnostic - #270
Draft
ProfSynapse wants to merge 8 commits into
Draft
ProfSynapse wants to merge 8 commits into
ProfSynapse wants to merge 8 commits into
Conversation
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Scope
Adds a constructor-only GRPO diagnostic under the existing derived-image qualification CLI. The probe derives exact four-row input from the normal plan/recipe/dataset, binds committed probe/helper/CLI source plus release/image/model inventory, and returns closed stage/category and class/signature hashes. The effectful mode uses one consumed claim, a private verified model copy, no network, one GPU, two CPUs, 8 GiB RAM, a 120-second workload wait, bounded logs, terminal inspection, and identity-checked container cleanup. It never explicitly calls training or generation.
The published Unsloth Zoo 2026.1.2 wheel requires an accelerator during import, so CPU cannot exercise the installed patched constructor. This is a diagnostic capability, not a fix for the consumed Modal run or training qualification.
Verification
3c5c179a0aca0e2ab9899c6a11cb6064761ec2df766e76c60899192093c89416.Review boundary
No Docker image pull/build, GPU/model download, Modal call, training, or old-resource cleanup occurred. The host has no usable GPU, and the derived image and retained Modal snapshot are not local. The reviewed GPU admission still needs exact image capture and an independently verified offline snapshot on the execution machine. See
docs/review/grpo-constructor-diagnostic.md. The old attempt remains consumed; no root cause is claimed.