jgrusewski 4fa9da9fc1 perf(cuda): enable TF32 Tensor Core math on all cuBLAS handles ...

DQN head (dqn.rs) and Mamba2 encoder (mamba2_block.rs) were forcing
CUBLAS_COMPUTE_32F — pure FP32 scalar, bypassing Tensor Cores entirely.
TF32 (19-bit mantissa) gives 2-3× SGEMM throughput on L40S/H100 with
negligible precision loss (well within RL gradient noise).

8+ SGEMMs per step now use Tensor Cores: DQN forward/backward (4) +
Mamba2 W_in/W_a/W_b/W_out projections (4+).

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

2026-05-27 20:19:45 +02:00

..

cuda

feat(cuda): re-enable C51 atom span EWMA anchored on clamp bounds

2026-05-27 15:04:20 +02:00

examples

fix(cuda): SAC co-tuning reads ACTION entropy, not policy entropy

2026-05-27 13:44:10 +02:00

src

perf(cuda): enable TF32 Tensor Core math on all cuBLAS handles

2026-05-27 20:19:45 +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

fix(cuda): SAC co-tuning reads ACTION entropy, not policy entropy

2026-05-27 13:44:10 +02:00

Cargo.toml

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

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