Skip to content
KernelIndex
Search⌘K
Sign in

Gaussian topk sparse activation

23 eligible runs
moe

Gaussian-based top-k sparse activation mechanism. Computes adaptive sparsity by calculating a threshold based on input's mean and standard deviation, scaled by the inverse CDF of a normal distribution at the target sparsity level. Applies ReLU with this dynamic cutoff to create learned sparse activations.

Fastest reported
17.9µsmean · 66% of SOL
RIAC as wellLicense unknown · unknown

Reported evidence · last observed 2026-08-22. Reported by source; not independently reproduced. No public source. License unknown.

No public sourceRun detail →

Current records

Workloadsuite of 12 cases · mean latency12 cases
seq_len
batch_size
intermediate_size
dtype
1283212288bf1613124096bf1621128192bf16293212288bf1644944096bf16512112288bf1654148192bf161024116384bf161024648192bf161163168192bf162048412288bf16819214096bf16
Aggregatessuite of 12 cases · mean latency
HardwareBest knownImplementationRuns
NVIDIA B200 · env 117.9 µsRIAC as well12NVIDIA B200 · env 219.2 µsDatabricks11

Not measured on H100 for this workload. Challenges →

Source-native comparison · GPU NVIDIA B200 · Workload suite of 12 cases · mean latency · Protocol SOL-ExecBench evaluation stack v1.1 · mean · 12 results · last observed 2026-08-22Record history →
#
Implementation
Latency
vs #1
Trust
Observed
1
RIAC as well
17.9µs66% SOL
1.00×
Reported · license unknown · no source
2026-08-22

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

no sourcelicense unknownno install recipeRun detail →
2
Recursive
18.5µs62% SOL
1.03×
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
Geometric
18.7µs61% SOL
1.04×
Reported · license unknown · no source
2026-08-12

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

no sourcelicense unknownno install recipeRun detail →
4
Databricks
18.9µs60% SOL
1.05×
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 →
5
ac4k
19.1µs60% SOL
1.06×
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
Amir M. Mir | SF Tensor
19.2µs58% SOL
1.07×
Reported · license unknown · no source
2026-08-03

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

no sourcelicense unknownno install recipeRun detail →
7
Amir M. Mir | SF Tensor
19.3µs58% SOL
1.08×
Reported · license unknown · no source
2026-07-31

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

no sourcelicense unknownno install recipeRun detail →
8
doubleAI
19.3µs59% 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 →
9
Hyra
19.5µs58% SOL
1.08×
Reported · license unknown · no source
2026-07-16

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

no sourcelicense unknownno install recipeRun detail →
10
Amir M. Mir | SF Tensor
19.6µs57% SOL
1.09×
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
Cursor
24.2µs42% SOL
1.35×
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 →
12
TH3CHARLie | TileGym
29.3µs35% SOL
1.63×
Reported · license unknown · no source
2026-07-22

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

no sourcelicense unknownno install recipeRun detail →

Implementations

Implementation
Runtime
Best latency
Evidence
Availability
RIAC as well
unknown
17.9µs
1.00×
Reported
license unknown · no source
ac4k
unknown
19.1µs
1.06×
Reported
license unknown · no source
doubleAI
unknown
19.3µs
1.08×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
19.2µs
1.07×
Reported
license unknown · no source
Hyra
unknown
19.5µs
1.08×
Reported
license unknown · no source
Argus
unknown
21.4µs
1.19×
Reported
license unknown · no source
Cursor
unknown
23.9µs
1.33×
Reported
license unknown · no source
TH3CHARLie | TileGym
unknown
29.3µs
1.63×
Reported
license unknown · no source
Show all 23 implementations ›
Bragaia
unknown
318.7µs
17.76×
Reported
license unknown · no source
doubleAI
unknown
19.3µs
1.08×
Reported
license unknown · no source
Databricks
unknown
19.2µs
1.07×
Reported
license unknown · no source
ac4k
unknown
19.3µs
1.08×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
19.3µs
1.08×
Reported
license unknown · no source
Recursive
unknown
19.2µs
1.07×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
19.6µs
1.09×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
19.7µs
1.10×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
21.4µs
1.19×
Reported
license unknown · no source
Cursor
unknown
24.2µs
1.35×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
32.3µs
1.80×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
63.7µs
3.55×
Reported
license unknown · no source
Recursive
unknown
18.5µs
1.03×
Reported
license unknown · no source
Geometric
unknown
18.7µs
1.04×
Reported
license unknown · no source
Databricks
unknown
18.9µs
1.05×
Reported
license unknown · no source

Semantics

Inputs and outputs
inputsbf16 [batch_size, seq_len, intermediate_size]
target_sparsityfp32 scalar
outputbf16 [batch_size, seq_len, intermediate_size]
Axes and behavior
seq_lenvariable
batch_sizevariable
intermediate_sizevariable
determinismunspecified
constraintsNo mutation or aliasing
Identity
alias053_gaussian_topk_sparse_activationmodelgemma-3n-e2bsha2565e52218b6d71…
Sources: NVIDIA SOL-ExecBench (2026-08-22)last observed 2026-08-22How records are decidedJSON