jgrusewski 514b04bee3 fix(rl): remove advantage normalization — sparse rewards diluted signal ...

With 5-6% of batch elements having real rewards per step, advantage
normalization (zero mean, unit var) turned 94% of the gradient into
noise. The /B gradient normalization already handles batch-size
invariance. Raw V-advantage bounded by reward_scale is sufficient.

Removed: normalize_advantages kernel (#if 0 in .cu), field, load,
both launch sites, q_logits_target_st_d (Q-advantage dead code).
ISV-driven TAU_MAX (slot 573, bootstrap 0.005) retained for future use.

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

2026-05-27 00:20:27 +02:00

..

cuda

fix(rl): remove advantage normalization — sparse rewards diluted signal

2026-05-27 00:20:27 +02:00

examples

fix(rl): wire BCE + aux losses into GPU loader path + event-based sync

2026-05-26 22:18:30 +02:00

src

fix(rl): remove advantage normalization — sparse rewards diluted signal

2026-05-27 00:20:27 +02:00

tests

perf(rl): fuse adamw_increment into adamw_step + tau into iqn_forward

2026-05-26 16:46:10 +02:00

build.rs

perf(rl): GPU-resident data loader — zero CPU per step

2026-05-26 20:51:06 +02:00

Cargo.toml

feat(rl): R6 — replace LobEnv with kernel-driven GPU-pure env step

2026-05-23 10:39:25 +02:00