v0.12.1: Unsloth 2026.9.2 / vLLM 0.28.0 images validated on Qwen3.8-27B and the full cookbook; DPO, GPU-reclaim and API-deploy fixes; training diagnostics - #81

Merged

nielsrolf merged 12 commits into

mainfrom

codex/training-validation

Sep 7, 2026

Commits on Sep 5, 2026

Fix DPO configuration and add paired training diagnostics
nielsrolf
committed
605fcf8
Merge remote-tracking branch 'origin/main' into codex/training-validation
nielsrolf
committed
8bcdd6f

Commits on Sep 6, 2026

Rebuild worker images on Unsloth 2026.9.2 and vLLM 0.28.0

nielsrolf
and
claude
committed
aa0b292
Training comparison: LR sweep, label audit, alpha-32 control and reports

nielsrolf
and
claude
committed
e5c22cc
Add built dashboard frontend assets for merged PRs

nielsrolf
and
claude
committed
b320f05
Fix DPO on models TRL classifies as vision models

nielsrolf
and
claude
committed
8235416
Release v0.12 images

nielsrolf
and
claude
committed
fa24eee
Record Qwen3.8-27B smoke tests and v0.12 image validation

nielsrolf
and
claude
committed
f5ac4c3
Integration runner: add --skip to bypass individual tests

nielsrolf
and
claude
committed
aba9d6b
Fix worker GPU reclaim aborting on small foreign holders; release v0.12.1

nielsrolf
and
claude
committed
33bfada
Fix ow.api.deploy() requiring OPENAI_API_KEY in the environment

nielsrolf
and
claude
committed
5aa149c
Record integration suite results and v0.12.1 validation

nielsrolf
and
claude
committed
07deb74