Skip to content
KernelIndex
Search⌘K

Vision patch merger spatial shuffle MLP

29 eligible runs
mlp

Vision patch merger that performs spatial shuffling to merge patches (spatial_merge_size=2), applies layer normalization, and processes through a 2-layer MLP with GELU activation. Converts fine-grained vision tokens into coarser representations by merging 2x2 spatial patches.

Fastest reported
144.2µsmean · 73% of SOL
RecursiveLicense unknown · unknown

Reported evidence · last observed 2026-07-10. Reported by source; not independently reproduced. No public source. License unknown.

No public sourceRun detail →

Current records

Workloadsuite of 15 cases · mean latencythree = 3 · hidden_size = 1536 · out_hidden_size = 3584 · spatial_merge_size = 2 · hidden_size_expanded = 614415 cases

Not measured on H100 for this workload. Challenges →

Source-native comparison · GPU NVIDIA B200 · Workload suite of 15 cases · mean latency · Protocol SOL-ExecBench evaluation stack v1.1 · mean · 15 results · last observed 2026-08-11Record history →
#
Implementation
Latency
vs #1
Trust
Observed
1
144.2µs73% SOL
1.00×
Reported · license unknown · no source
2026-07-10

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

no sourcelicense unknownno install recipeRun detail →
2
147.5µs72% SOL
1.02×
Reported · license unknown · no source
2026-07-10

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

no sourcelicense unknownno install recipeRun detail →
3
155.2µs70% SOL
1.08×
Reported · license unknown · no source
2026-07-10

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

no sourcelicense unknownno install recipeRun detail →
4
156.2µs70% SOL
1.08×
Reported · license unknown · no source
2026-07-28

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

no sourcelicense unknownno install recipeRun detail →
5
162.5µs67% SOL
1.13×
Reported · license unknown · no source
2026-07-10

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

no sourcelicense unknownno install recipeRun detail →
6
164.5µs67% SOL
1.14×
Reported · license unknown · no source
2026-07-19

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

no sourcelicense unknownno install recipeRun detail →
7
166.9µs66% SOL
1.16×
Reported · license unknown · no source
2026-08-11

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

no sourcelicense unknownno install recipeRun detail →
8
167.0µs66% SOL
1.16×
Reported · license unknown · no source
2026-08-07

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

no sourcelicense unknownno install recipeRun detail →
9
168.7µs66% SOL
1.17×
Reported · license unknown · no source
2026-07-12

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

no sourcelicense unknownno install recipeRun detail →
10
169.4µs66% SOL
1.17×
Reported · license unknown · no source
2026-07-10

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

no sourcelicense unknownno install recipeRun detail →
11
175.5µs64% SOL
1.22×
Reported · license unknown · no source
2026-08-06

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

no sourcelicense unknownno install recipeRun detail →
12
213.0µs55% SOL
1.48×
Reported · license unknown · no source
2026-07-10

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

no sourcelicense unknownno install recipeRun detail →
13
230.8µs53% SOL
1.60×
Reported · license unknown · no source
2026-07-17

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

no sourcelicense unknownno install recipeRun detail →
14
262.4µs49% SOL
1.82×
Reported · license unknown · no source
2026-07-23

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

no sourcelicense unknownno install recipeRun detail →
15
267.5µs48% SOL
1.85×
Reported · license unknown · no source
2026-07-24

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
141.9µs
1.03×
Reported
license unknown · no source
unknown
147.1µs
1.06×
Reported
license unknown · no source
unknown
156.2µs
1.13×
Reported
license unknown · no source
unknown
164.5µs
1.19×
Reported
license unknown · no source
unknown
165.8µs
1.20×
Reported
license unknown · no source
unknown
166.9µs
1.21×
Reported
license unknown · no source
unknown
182.0µs
1.31×
Reported
license unknown · no source
unknown
199.3µs
1.44×
Reported
license unknown · no source
Show all 29 implementations ›
unknown
230.8µs
1.67×
Reported
license unknown · no source
unknown
301.2µs
2.18×
Reported
license unknown · no source
unknown
147.5µs
1.07×
Reported
license unknown · no source
unknown
155.2µs
1.12×
Reported
license unknown · no source
unknown
163.3µs
1.18×
Reported
license unknown · no source
unknown
162.5µs
1.17×
Reported
license unknown · no source
unknown
165.8µs
1.20×
Reported
license unknown · no source
unknown
167.0µs
1.21×
Reported
license unknown · no source
unknown
168.7µs
1.22×
Reported
license unknown · no source
unknown
169.4µs
1.22×
Reported
license unknown · no source
unknown
175.5µs
1.27×
Reported
license unknown · no source
unknown
177.7µs
1.28×
Reported
license unknown · no source
unknown
213.0µs
1.54×
Reported
license unknown · no source
unknown
230.0µs
1.66×
Reported
license unknown · no source
unknown
230.9µs
1.67×
Reported
license unknown · no source
unknown
260.3µs
1.88×
Reported
license unknown · no source
unknown
262.4µs
1.90×
Reported
license unknown · no source
unknown
267.5µs
1.93×
Reported
license unknown · no source
unknown
138.4µs
1.00×
Reported
license unknown · no source
unknown
144.2µs
1.04×
Reported
license unknown · no source
unknown
146.0µs
1.06×
Reported
license unknown · no source

Semantics

Inputs and outputs
hiddenbf16 [num_patches, hidden_size]
grid_thwint64 [num_grids, three]
ln_weightbf16 [hidden_size]
ln_biasbf16 [hidden_size]
fc1_weightbf16 [hidden_size_expanded, hidden_size_expanded]
fc1_biasbf16 [hidden_size_expanded]
fc2_weightbf16 [out_hidden_size, hidden_size_expanded]
fc2_biasbf16 [out_hidden_size]
epsfp32 scalar
outputbf16 [num_merged_patches, out_hidden_size]
Axes and behavior
threeconstant = 3
num_gridsvariable
hidden_sizeconstant = 1536
num_patchesvariable
out_hidden_sizeconstant = 3584
num_merged_patchesvariable
spatial_merge_sizeconstant = 2
hidden_size_expandedconstant = 6144
determinismunspecified
constraintsNo mutation or aliasing
Identity
alias020_vision_patch_merger_spatial_shuffle_mlpmodelqwen3-omni-30b-a3b-instructsha2568d5db089ce07…
Sources: NVIDIA SOL-ExecBench (2026-08-11)last observed 2026-08-11How records are decidedJSON