Vision language cross attention fusion
Cross-attention fusion layer for Qwen3-VL multimodal processing. Performs cross-attention between language queries (with 1D RoPE) and vision key-values (with 3D RoPE for temporal-height-width positions). Uses Grouped Query Attention (GQA) with 32 query heads and 8 key-value heads.
Reported evidence · last observed 2026-08-15. Reported by source; not independently reproduced. No public source. License unknown.
Current records
Workloadsuite of 16 cases · mean latencythw_dim = 3 · head_dim = 128 · q_proj_dim = 4096 · hidden_size = 4096 · kv_proj_dim = 1024 · num_attention_heads = 32 · num_key_value_heads = 816 cases
Not measured on H100 for this workload. Challenges →
1286.2µs65% SOL1.00×Reported · license unknown · no source2026-08-15
Measured exactly what you asked. Reported by source; not independently reproduced.
2303.9µs63% SOL1.06×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
3305.5µs63% SOL1.07×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
4309.4µs62% SOL1.08×Reported · license unknown · no source2026-07-17
Measured exactly what you asked. Reported by source; not independently reproduced.
5327.2µs60% SOL1.14×Reported · license unknown · no source2026-08-05
Measured exactly what you asked. Reported by source; not independently reproduced.
6330.8µs59% SOL1.16×Reported · license unknown · no source2026-07-17
Measured exactly what you asked. Reported by source; not independently reproduced.
7331.4µs60% SOL1.16×Reported · license unknown · no source2026-08-20
Measured exactly what you asked. Reported by source; not independently reproduced.
8331.8µs60% SOL1.16×Reported · license unknown · no source2026-08-18
Measured exactly what you asked. Reported by source; not independently reproduced.
9333.2µs59% SOL1.16×Reported · license unknown · no source2026-07-11
Measured exactly what you asked. Reported by source; not independently reproduced.
10339.8µs58% SOL1.19×Reported · license unknown · no source2026-08-17
Measured exactly what you asked. Reported by source; not independently reproduced.
11340.6µs58% SOL1.19×Reported · license unknown · no source2026-08-16
Measured exactly what you asked. Reported by source; not independently reproduced.
12405.6µs52% SOL1.42×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
13430.3µs51% SOL1.50×Reported · license unknown · no source2026-08-20
Measured exactly what you asked. Reported by source; not independently reproduced.
14431.3µs50% SOL1.51×Reported · license unknown · no source2026-08-13
Measured exactly what you asked. Reported by source; not independently reproduced.
15450.5µs49% SOL1.57×Reported · license unknown · no source2026-08-18
Measured exactly what you asked. Reported by source; not independently reproduced.
16467.7µs46% SOL1.63×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.