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

..

backtesting

…

common

…

config

…

ctrader-openapi

…

data

fix(loader): soft-validate FrontMonth symbol when no streaming SymbolMapping

2026-05-22 18:08:36 +02:00

database

…

market-data

…

ml

fix(ml): migrate test uploads from clone_htod to mapped-pinned

2026-05-22 20:39:32 +02:00

ml-alpha

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

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

ml-asset-selection

…

ml-backtesting

perf(lobsim): convert step_fill + step_pnl_track to raw_launch

2026-05-27 11:25:48 +02:00

ml-checkpoint

…

ml-core

fix(ml-core): deterministic GPU weight init via scoped_init_seed

2026-05-19 01:19:13 +02:00

ml-data

…

ml-data-validation

…

ml-dqn

feat(sp22-vnext): Phase B4b-2 — replay-buffer scatter + trainer setter

2026-05-14 02:13:05 +02:00

ml-ensemble

…

ml-explainability

…

ml-features

fix(ml-features): imbalance bar cache on persistent PVC, not /tmp

2026-05-24 23:11:03 +02:00

ml-hyperopt

…

ml-labeling

…

ml-observability

…

ml-paper-trading

…

ml-ppo

…

ml-regime

…

ml-regime-detection

…

ml-risk

…

ml-security

…

ml-stress-testing

…

ml-supervised

…

ml-universe

…

ml-validation

…

model_loader

…

risk

…

risk-data

…

storage

…

trading_engine

…

trading-data

…

training_uploader

…