jgrusewski 148454d37d feat(cuda): curriculum weights E8 + adversarial regime boost kernels ...

rl_curriculum_weights: per-segment Sharpe → z-score → softmax difficulty
weights. Harder segments sampled more. Block tree-reduce, no atomics.

rl_adversarial_boost: multiply PER priority by boost factor for
negative-reward transitions. Self-regulating — fewer losses → fewer
boosts. ISV-driven threshold + boost magnitude.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

2026-05-26 12:00:21 +02:00

6.7 KiB

Raw Blame History

View Raw