jgrusewski 51a1b1b89b spec: unified cuBLAS training graph — IQL/IQN/attention rewrite + graph unification ...

Single cudarc-captured graph replaces 6 separate graphs.
IQL/IQN/Attention rewritten from per-sample kernels to batched cuBLAS.
Temporal ops inside unified graph. Target: <80s epochs from 685s.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

2026-04-17 22:57:14 +02:00

..

infra

feat(infra): GPU training launcher with local/cloud routing

2026-02-23 11:19:58 +01:00

plans

feat(dqn): enable Branching DQN with 45 factored actions (5×3×3)

2026-03-11 22:00:13 +01:00

superpowers

spec: unified cuBLAS training graph — IQL/IQN/attention rewrite + graph unification

2026-04-17 22:57:14 +02:00