SolitaryThinker 76129f5a89 [perf]: add bounded H3 windows and agreed GB200 launch plans

2026-09-06 20:41:05 +00:00

..

attention

[kernel] sm_100a CUDA block-sparse VSA forward (Blackwell), 64- and 128-token blocks (#1719)

2026-08-20 23:16:25 -05:00

comm

[perf]: add bounded H3 windows and agreed GB200 launch plans

2026-09-06 20:41:05 +00:00

turbodiffusion

[infra] Add DGX Spark and multi-architecture CUDA support (#1447)

2026-07-03 13:41:43 -07:00

common_extension.cpp

[perf] Add fused NVLink all-to-all for Ulysses (#1740)

2026-08-22 23:06:24 -07:00