chore(release): 0.57.0 — GPU QLoRA fine-tuning works + runnable merge + 3 enforced beats#2256
Merged
Conversation
…e) + runnable merge + 3 enforced beats The release where apr finetune -m qlora goes from deadlocks-on-first-forward to trains-the-correct-model on RTX 4090 (verified on GB10 Blackwell): a four-defect cascade (#2249 self-deadlock, #2250 loss-window clamp, #2251 cuBLAS cross-stream NaN race, #2252 wrong-model quartet), each with a mutation-verified falsifier + pv-validated contract. Plus apr finetune --merge now produces a directly-runnable .apr (#2254), and three already-measured beats that claimed CI enforcement but ran nowhere are now gated per-PR (#2253). Version bump 0.56.0 -> 0.57.0 across root + 32 crate manifests (incl. two-component apr-format sibling dep) + Cargo.lock regen. CHANGELOG entry added. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
v0.57.0 — GPU QLoRA fine-tuning actually works
Version bump
0.56.0 → 0.57.0across root + 32 crate manifests (incl. the two-componentapr-formatsibling dep) +Cargo.lockregen + CHANGELOG.Headline
apr finetune -m qlorawent from deadlocks on the first transformer block to trains the correct model on an RTX 4090 — cross-silicon verified on a GB10 Blackwell (sm_121). A four-defect cascade, each already merged with a mutation-verified RED→GREEN falsifier +pv-validated contract:--max-seq-len(was a hardcoded-512 no-op-epoch clamp)Plus:
apr finetune --mergeproduces a directly-runnable.apr(fail-closed gate + duplicate-field poison + GGUF↔HF adapter naming)Also this cycle: beat-speed-nightly restored to GREEN (corrupt uv cache on the runner) — ~11 speed beats across P1-P4 un-blinded.
All content already merged to main via the individual PRs above; this PR is the version-bump + CHANGELOG cut. Full workspace builds clean at 0.57.0.
🤖 Generated with Claude Code