Skip to content
KernelIndex
Search⌘K

Conformer relative position attention

26 eligible runs
audio

Conformer attention with Shaw's relative positional embeddings. Processes sequences in fixed-size context blocks (512 tokens), computes scaled dot-product attention within each block, and adds relative positional bias via learned embeddings. Handles dynamic padding and masking for incomplete blocks.

Fastest reported
103.6µsmean · 73% of SOL
RIAC as wellLicense unknown · unknown

Reported evidence · last observed 2026-08-17. Reported by source; not independently reproduced. No public source. License unknown.

No public sourceRun detail →

Current records

Workloadsuite of 16 cases · mean latencydim_head = 128 · inner_dim = 1024 · num_heads = 8 · hidden_dim = 1024 · max_pos_emb = 512 · num_rel_pos = 1025 · context_size = 512 · inner_dim_x2 = 204816 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 v1.1 · mean · 18 results · last observed 2026-08-21Record history →
#
Implementation
Latency
vs #1
Trust
Observed
1
103.6µs73% SOL
1.00×
Reported · license unknown · no source
2026-08-17

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

no sourcelicense unknownno install recipeRun detail →
2
111.2µs71% SOL
1.07×
Reported · license unknown · no source
2026-07-15

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

no sourcelicense unknownno install recipeRun detail →
3
129.8µs67% SOL
1.25×
Reported · license unknown · no source
2026-08-21

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

no sourcelicense unknownno install recipeRun detail →
4
135.4µs66% SOL
1.31×
Reported · license unknown · no source
2026-08-14

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

no sourcelicense unknownno install recipeRun detail →
5
136.4µs66% SOL
1.32×
Reported · license unknown · no source
2026-08-04

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

no sourcelicense unknownno install recipeRun detail →
6
136.9µs66% SOL
1.32×
Reported · license unknown · no source
2026-08-17

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

no sourcelicense unknownno install recipeRun detail →
7
139.0µs65% SOL
1.34×
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 →
8
160.3µs62% SOL
1.55×
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
166.3µs61% 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 →
10
171.0µs60% SOL
1.65×
Reported · license unknown · no source
2026-08-17

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

no sourcelicense unknownno install recipeRun detail →
11
203.8µs55% SOL
1.97×
Reported · license unknown · no source
2026-07-25

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

no sourcelicense unknownno install recipeRun detail →
12
211.0µs54% SOL
2.04×
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
229.1µs52% SOL
2.21×
Reported · license unknown · no source
2026-08-15

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

no sourcelicense unknownno install recipeRun detail →
14
267.9µs48% SOL
2.59×
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 →
15
330.5µs42% SOL
3.19×
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 →
16
425.7µs36% SOL
4.11×
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 →
17
510.2µs32% SOL
4.92×
Reported · license unknown · no source
2026-08-14

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

no sourcelicense unknownno install recipeRun detail →
18
639.7µs29% SOL
6.17×
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 →

Implementations

Implementation
Runtime
Best latency
Evidence
Availability
unknown
103.6µs
1.00×
Reported
license unknown · no source
unknown
129.8µs
1.25×
Reported
license unknown · no source
unknown
135.4µs
1.31×
Reported
license unknown · no source
unknown
136.4µs
1.32×
Reported
license unknown · no source
unknown
156.5µs
1.51×
Reported
license unknown · no source
unknown
166.3µs
1.60×
Reported
license unknown · no source
unknown
252.4µs
2.44×
Reported
license unknown · no source
unknown
374.8µs
3.62×
Reported
license unknown · no source
Show all 26 implementations ›
unknown
136.9µs
1.32×
Reported
license unknown · no source
unknown
139.0µs
1.34×
Reported
license unknown · no source
unknown
160.3µs
1.55×
Reported
license unknown · no source
unknown
171.0µs
1.65×
Reported
license unknown · no source
unknown
203.8µs
1.97×
Reported
license unknown · no source
unknown
211.0µs
2.04×
Reported
license unknown · no source
unknown
229.1µs
2.21×
Reported
license unknown · no source
unknown
242.2µs
2.34×
Reported
license unknown · no source
unknown
267.9µs
2.59×
Reported
license unknown · no source
unknown
330.5µs
3.19×
Reported
license unknown · no source
unknown
420.9µs
4.06×
Reported
license unknown · no source
unknown
425.7µs
4.11×
Reported
license unknown · no source
unknown
510.2µs
4.92×
Reported
license unknown · no source
unknown
639.7µs
6.17×
Reported
license unknown · no source
unknown
1.20ms
11.61×
Reported
license unknown · no source
unknown
111.2µs
1.07×
Reported
license unknown · no source
unknown
130.8µs
1.26×
Reported
license unknown · no source
unknown
133.7µs
1.29×
Reported
license unknown · no source

Semantics

Inputs and outputs
hidden_statesbf16 [batch_size, seq_len, hidden_dim]
pre_norm_weightbf16 [hidden_dim]
pre_norm_biasbf16 [hidden_dim]
to_q_weightbf16 [inner_dim, hidden_dim]
to_kv_weightbf16 [inner_dim_x2, hidden_dim]
to_out_weightbf16 [hidden_dim, inner_dim]
to_out_biasbf16 [hidden_dim]
rel_pos_emb_weightbf16 [num_rel_pos, dim_head]
scalefp32 scalar
outputbf16 [batch_size, seq_len, hidden_dim]
Axes and behavior
seq_lenvariable
dim_headconstant = 128
inner_dimconstant = 1024
num_headsconstant = 8
batch_sizevariable
hidden_dimconstant = 1024
max_pos_embconstant = 512
num_rel_posconstant = 1025
context_sizeconstant = 512
inner_dim_x2constant = 2048
determinismunspecified
constraintsNo mutation or aliasing
Identity
alias046_conformer_relative_position_attentionmodelgranite-speech-3-3-8bsha256a9c63608df8e…
Sources: NVIDIA SOL-ExecBench (2026-08-21)last observed 2026-08-21How records are decidedJSON