Fix the bf16 medium DiT: bake RoPE's tables into the graph (clipping 3.112% -> 0.014%) 59bd967 verified cortexelus commited on 23 days ago
Rebuild medium DiT fp16-mixed engines (sm_90 + new sm_120): attention now fuses, 4.3x faster (#4) 06debc3 cortexelus commited on 23 days ago
