Skip to content
KernelIndex
Search⌘K

Hybrid attention mask preparation

31 eligible runs
other

Hybrid causal mask preparation system that creates separate attention masks for full attention and sliding window attention layers. Creates two distinct 4D causal masks: standard causal masking for full attention layers and sliding window causal masking for SWA layers.

Fastest reported
954nsmean · 94% of SOL
doubleAILicense unknown · unknown

Reported evidence · last observed 2026-06-25. Reported by source; not independently reproduced. No public source. License unknown.

No public sourceRun detail →

Current records

Workloadsuite of 12 cases · mean latencysliding_window = 128 · num_attention_heads = 64 · swa_num_attention_heads = 6412 cases

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 v1.0 · mean · 16 results · last observed 2026-08-03Record history →
#
Implementation
Latency
vs #1
Trust
Observed
1
954ns94% SOL
1.00×
Reported · license unknown · no source
2026-06-25

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

no sourcelicense unknownno install recipeRun detail →
2
1.17µs92% SOL
1.22×
Reported · license unknown · no source
2026-07-02

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

no sourcelicense unknownno install recipeRun detail →
3
1.94µs88% SOL
2.04×
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 →
4
1.98µs87% SOL
2.08×
Reported · license unknown · no source
2026-06-06

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

no sourcelicense unknownno install recipeRun detail →
5
2.15µs86% SOL
2.25×
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 →
6
2.27µs85% SOL
2.38×
Reported · license unknown · no source
2026-07-08

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

no sourcelicense unknownno install recipeRun detail →
7
2.56µs83% SOL
2.69×
Reported · license unknown · no source
2026-07-03

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

no sourcelicense unknownno install recipeRun detail →
8
2.66µs82% SOL
2.79×
Reported · license unknown · no source
2026-07-08

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

no sourcelicense unknownno install recipeRun detail →
9
5.91µs64% SOL
6.20×
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 →
10
25.1µs30% SOL
26.35×
Reported · license unknown · no source
2026-05-28

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

no sourcelicense unknownno install recipeRun detail →
11
73.2µs28% SOL
76.69×
Reported · license unknown · no source
2026-05-30

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

no sourcelicense unknownno install recipeRun detail →
12
96.4µs27% SOL
101.00×
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 →
13
99.3µs27% SOL
104.09×
Reported · license unknown · no source
2026-05-28

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

no sourcelicense unknownno install recipeRun detail →
14
176.2µs20% SOL
184.69×
Reported · license unknown · no source
2026-05-16

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

no sourcelicense unknownno install recipeRun detail →
15
184.9µs24% SOL
193.80×
Reported · license unknown · no source
2026-05-16

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

no sourcelicense unknownno install recipeRun detail →
16
346.9µs12% SOL
363.60×
Reported · license unknown · no source
2026-04-21

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
1.17µs
1.22×
Reported
license unknown · no source
unknown
1.21µs
1.27×
Reported
license unknown · no source
unknown
954ns
1.00×
Reported
license unknown · no source
unknown
1.79µs
1.88×
Reported
license unknown · no source
unknown
1.94µs
2.04×
Reported
license unknown · no source
unknown
1.98µs
2.08×
Reported
license unknown · no source
unknown
2.04µs
2.14×
Reported
license unknown · no source
unknown
2.27µs
2.38×
Reported
license unknown · no source
Show all 31 implementations ›
unknown
3.05µs
3.20×
Reported
license unknown · no source
unknown
5.83µs
6.11×
Reported
license unknown · no source
unknown
96.4µs
101.00×
Reported
license unknown · no source
unknown
1.30µs
1.36×
Reported
license unknown · no source
unknown
1.94µs
2.03×
Reported
license unknown · no source
unknown
2.08µs
2.18×
Reported
license unknown · no source
unknown
2.15µs
2.25×
Reported
license unknown · no source
unknown
2.42µs
2.54×
Reported
license unknown · no source
unknown
2.49µs
2.61×
Reported
license unknown · no source
unknown
2.56µs
2.69×
Reported
license unknown · no source
unknown
2.66µs
2.79×
Reported
license unknown · no source
unknown
2.69µs
2.82×
Reported
license unknown · no source
unknown
2.74µs
2.87×
Reported
license unknown · no source
unknown
5.91µs
6.20×
Reported
license unknown · no source
unknown
40.4µs
42.36×
Reported
license unknown · no source
unknown
42.6µs
44.61×
Reported
license unknown · no source
unknown
25.1µs
26.35×
Reported
license unknown · no source
unknown
73.2µs
76.69×
Reported
license unknown · no source
unknown
99.3µs
104.09×
Reported
license unknown · no source
unknown
184.9µs
193.80×
Reported
license unknown · no source
unknown
176.2µs
184.69×
Reported
license unknown · no source
unknown
346.9µs
363.60×
Reported
license unknown · no source
unknown
955ns
1.00×
Reported
license unknown · no source

Semantics

Inputs and outputs
batch_size_scalarint64 scalar
seq_length_scalarint64 scalar
past_key_values_length_scalarint64 scalar
full_attention_maskbool [batch_size, num_attention_heads, seq_length, total_length]
sliding_window_attention_maskbool [batch_size, swa_num_attention_heads, seq_length, total_length]
Axes and behavior
batch_sizevariable
seq_lengthvariable
total_lengthtotal_length = seq_length + past_key_values_length
sliding_windowconstant = 128
num_attention_headsconstant = 64
past_key_values_lengthvariable
swa_num_attention_headsconstant = 64
determinismunspecified
constraintsNo mutation or aliasing
Identity
alias028_hybrid_attention_mask_preparationmodelmimo-v2-flashsha256dc8fb6429f3a…
Sources: NVIDIA SOL-ExecBench (2026-08-03)last observed 2026-08-03How records are decidedJSON