jgrusewski 6674007952 feat(rl): wire GPU PER — push/sample/update/rebuild all device-side ...

Replaces todo!() stubs with kernel launches. GpuReplayBuffer field,
4 cubin loads, full per-step pipeline:
1. rl_per_push after reward extraction (n-step + circular write)
2. rl_per_sample in K-loop (tree walk + gather to sampled_*_d)
3. rl_per_update_priority + tree_rebuild after step_synthetic

Zero host-side PER operations. Per feedback_cpu_is_read_only.

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

2026-05-26 00:36:34 +02:00

..

cuda

feat(cuda): GPU-resident PER kernels — push, sample, update, tree_rebuild

2026-05-26 00:25:31 +02:00

examples

refactor(rl): delete host-side PER + diag-every band-aid (greenfield prep)

2026-05-26 00:18:11 +02:00

src

feat(rl): wire GPU PER — push/sample/update/rebuild all device-side

2026-05-26 00:36:34 +02:00

tests

test(rl): GPU oracle tests for IQN + NoisyNet + PRNG advance fix

2026-05-25 22:37:40 +02:00

build.rs

feat(cuda): GPU-resident PER kernels — push, sample, update, tree_rebuild

2026-05-26 00:25:31 +02:00

Cargo.toml

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

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