Video spatial attention with RoPE 3d
Video Spatial Attention with 3D Rotary Position Embedding. Applies multi-head self-attention to video tokens with 3D RoPE encoding spatial (height, width) and temporal (frame) positions. Critical for text-to-video generation where spatial coherence and temporal awareness are essential.
Reported evidence · last observed 2026-08-19. Reported by source; not independently reproduced. No public source. License unknown.
Current records
Workloadsuite of 16 cases · mean latencyhead_dim = 64 · qkv_hidden = 3072 · hidden_size = 1024 · half_head_dim = 32 · num_attention_heads = 1616 cases
Not measured on H100 for this workload. Challenges →
1132.6µs89% SOL1.00×Reported · license unknown · no source2026-08-19
Measured exactly what you asked. Reported by source; not independently reproduced.
2142.1µs88% SOL1.07×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
3149.3µs88% SOL1.13×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
4149.7µs88% SOL1.13×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
5154.8µs87% SOL1.17×Reported · license unknown · no source2026-07-15
Measured exactly what you asked. Reported by source; not independently reproduced.
6155.8µs87% SOL1.17×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
7166.3µs86% SOL1.25×Reported · license unknown · no source2026-07-12
Measured exactly what you asked. Reported by source; not independently reproduced.
8169.5µs86% SOL1.28×Reported · license unknown · no source2026-08-22
Measured exactly what you asked. Reported by source; not independently reproduced.
9171.1µs86% SOL1.29×Reported · license unknown · no source2026-08-18
Measured exactly what you asked. Reported by source; not independently reproduced.
10172.1µs86% SOL1.30×Reported · license unknown · no source2026-08-22
Measured exactly what you asked. Reported by source; not independently reproduced.
11177.1µs85% SOL1.34×Reported · license unknown · no source2026-08-23
Measured exactly what you asked. Reported by source; not independently reproduced.
12180.4µs85% SOL1.36×Reported · license unknown · no source2026-08-18
Measured exactly what you asked. Reported by source; not independently reproduced.
13181.6µs85% SOL1.37×Reported · license unknown · no source2026-08-22
Measured exactly what you asked. Reported by source; not independently reproduced.
14183.1µs85% SOL1.38×Reported · license unknown · no source2026-08-15
Measured exactly what you asked. Reported by source; not independently reproduced.
15195.5µs83% SOL1.47×Reported · license unknown · no source2026-07-20
Measured exactly what you asked. Reported by source; not independently reproduced.
16196.9µs83% SOL1.48×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
17204.3µs82% SOL1.54×Reported · license unknown · no source2026-07-27
Measured exactly what you asked. Reported by source; not independently reproduced.
18732.1µs56% SOL5.52×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
191.68ms35% SOL12.69×Reported · license unknown · no source2026-07-23
Measured exactly what you asked. Reported by source; not independently reproduced.