foxhunt

Files

jgrusewski 1c0c246ddf perf(rl): decouple diagnostic JSON writer to background thread

The 94KB per-step JSON record (130 fields, per-batch unit arrays)
took ~150ms CPU to build+serialize, blocking the GPU pipeline that
only needs ~13ms. Training was CPU-bound at 6.2 sps.

New architecture: training loop snapshots DiagFrame data (~0.1ms
memcpy from mapped-pinned to owned Vecs), sends via non-blocking
try_send on sync_channel(1). Background writer thread builds JSON
and writes to BufWriter at its own pace. Drops frames under
backpressure — acceptable for diagnostics.

Training loop per-step: 13ms GPU + 0.1ms snapshot = ~13.1ms.
Expected: ~50-77 sps at b=1024 on L40S (was 6.2).

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

2026-05-27 21:30:41 +02:00

cuda

perf(cuda): fuse 5 label gathers into sample_and_gather — 95.6% GPU time eliminated

2026-05-27 21:14:26 +02:00

examples

perf(rl): decouple diagnostic JSON writer to background thread

2026-05-27 21:30:41 +02:00

src

perf(cuda): fuse 5 label gathers into sample_and_gather — 95.6% GPU time eliminated

2026-05-27 21:14:26 +02:00

tests

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