fix(rl): GRPO training metrics - #1264

Draft

albcui wants to merge 7 commits into

aalgo-418/solufrom

grpo-training-metrics/albcui

Commits on Aug 12, 2026

fix(rl): report the GRPO metrics that say whether training is working
albcui
committed
d3a404d
test(rl): stop the chat_template test depending on ambient transformers
albcui
committed
9961fba
refactor(rl): fold the RL progress callback onto the shared implementation
albcui
committed
5a39483
fix(customization): stop every job from erasing its own loss curve
albcui
committed
78d8463
fix(rl): actually invoke the teardown that flushes the final train step
albcui
committed
c6341f8
fix(rl): stop double-counting the training step
albcui
committed
5366a3a
chore(rl): tighten four loose ends from review
albcui
committed
11ee9ec