Skip to content
KernelIndex
Search⌘K

VAE attention block with GroupNorm

25 eligible runs
rmsnorm

VAE self-attention block with group normalization. Applies group norm, flattens spatial dimensions (B,C,H,W) to (B,H*W,C), computes single-head self-attention over spatial positions, reshapes back, and adds residual. Critical for capturing long-range spatial dependencies in VAE encoder/decoder.

Fastest reported
634.5µsmean · 74% of SOL
RecursiveLicense unknown · unknown

Reported evidence · last observed 2026-06-07. Reported by source; not independently reproduced. No public source. License unknown.

No public sourceRun detail →

Current records

Workloadsuite of 16 cases · mean latencyseq_len = 4096 · channels = 512 · num_groups = 32 · spatial_size = 6416 cases

Not measured on H100 for this workload. Challenges →

Source-native comparison · GPU NVIDIA B200 · Workload suite of 16 cases · mean latency · Protocol SOL-ExecBench evaluation stack v1.0 · mean · 13 results · last observed 2026-08-18Record history →
#
Implementation
Latency
vs #1
Trust
Observed
1
634.5µs74% SOL
1.00×
Reported · license unknown · no source
2026-06-07

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
2
642.2µs73% SOL
1.01×
Reported · license unknown · no source
2026-06-29

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
3
646.8µs73% SOL
1.02×
Reported · license unknown · no source
2026-07-09

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
4
649.0µs73% SOL
1.02×
Reported · license unknown · no source
2026-06-26

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
5
651.5µs73% SOL
1.03×
Reported · license unknown · no source
2026-06-29

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
6
664.5µs72% SOL
1.05×
Reported · license unknown · no source
2026-07-05

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
7
696.3µs70% SOL
1.10×
Reported · license unknown · no source
2026-05-20

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
8
990.0µs58% SOL
1.56×
Reported · license unknown · no source
2026-06-11

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
9
1.31ms49% SOL
2.07×
Reported · license unknown · no source
2026-06-01

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
10
1.42ms47% SOL
2.24×
Reported · license unknown · no source
2026-04-25

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
11
1.74ms41% SOL
2.74×
Reported · license unknown · no source
2026-07-01

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
12
2.08ms36% SOL
3.27×
Reported · license unknown · no source
2026-04-14

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
13
10.9ms9% SOL
17.25×
Reported · license unknown · no source
2026-04-23

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →

Implementations

Implementation
Runtime
Best latency
Evidence
Availability
unknown
553.9µs
1.00×
Reported
license unknown · no source
unknown
642.2µs
1.16×
Reported
license unknown · no source
unknown
646.8µs
1.17×
Reported
license unknown · no source
unknown
649.0µs
1.17×
Reported
license unknown · no source
unknown
651.5µs
1.18×
Reported
license unknown · no source
unknown
693.5µs
1.25×
Reported
license unknown · no source
unknown
947.8µs
1.71×
Reported
license unknown · no source
unknown
1.74ms
3.13×
Reported
license unknown · no source
Show all 25 implementations ›
unknown
2.08ms
3.75×
Reported
license unknown · no source
unknown
640.6µs
1.16×
Reported
license unknown · no source
unknown
664.5µs
1.20×
Reported
license unknown · no source
unknown
666.4µs
1.20×
Reported
license unknown · no source
unknown
668.6µs
1.21×
Reported
license unknown · no source
unknown
669.6µs
1.21×
Reported
license unknown · no source
unknown
676.0µs
1.22×
Reported
license unknown · no source
unknown
680.4µs
1.23×
Reported
license unknown · no source
unknown
696.3µs
1.26×
Reported
license unknown · no source
unknown
990.0µs
1.79×
Reported
license unknown · no source
unknown
1.31ms
2.37×
Reported
license unknown · no source
unknown
1.42ms
2.57×
Reported
license unknown · no source
unknown
1.46ms
2.64×
Reported
license unknown · no source
unknown
10.9ms
19.75×
Reported
license unknown · no source
unknown
625.4µs
1.13×
Reported
license unknown · no source
unknown
634.5µs
1.15×
Reported
license unknown · no source
unknown
640.1µs
1.16×
Reported
license unknown · no source

Semantics

Inputs and outputs
xfp32 [batch_size, channels, spatial_size, spatial_size]
group_norm_weightfp32 [channels]
group_norm_biasfp32 [channels]
query_weightfp32 [channels, channels]
query_biasfp32 [channels]
key_weightfp32 [channels, channels]
key_biasfp32 [channels]
value_weightfp32 [channels, channels]
value_biasfp32 [channels]
proj_out_weightfp32 [channels, channels]
proj_out_biasfp32 [channels]
epsfp32 scalar
outputfp32 [batch_size, channels, spatial_size, spatial_size]
Axes and behavior
seq_lenseq_len = spatial_size * spatial_size
channelsconstant = 512
batch_sizevariable
num_groupsconstant = 32
spatial_sizeconstant = 64
determinismunspecified
constraintsNo mutation or aliasing
Identity
alias089_vae_attention_block_with_groupnormmodelhunyuanimage-2-1sha2562a79c795f17f…
Sources: NVIDIA SOL-ExecBench (2026-08-18)last observed 2026-08-18How records are decidedJSON