Skip to content
KernelIndex
Search⌘K
Sign in

Attention softmax dropout value matmul backward

128 eligible runs
attention

Backward pass for attention softmax, dropout, and value matmul. Computes gradients through: transpose -> batched matmul -> dropout -> softmax -> GQA head expansion.

Fastest reported
238.7µsmean · 79% of SOL
RIAC as wellLicense unknown · unknown

Reported evidence · last observed 2026-08-12. Reported by source; not independently reproduced. No public source. License unknown.

No public sourceRun detail →

Current records

Workloadsuite of 16 cases · mean latencyhead_dim = 128 · num_attention_heads = 80 · num_key_value_heads = 8 · num_key_value_groups = 1016 cases
seq_len_kv
seq_len_q
batch_size
dtype
1281288bf1612812832bf1612812864bf1625612816bf162562564bf162562568bf162932114bf164493738bf165122562bf1651225616bf165125121bf1651251232bf1677369132bf16102410241bf16204810244bf16409640961bf16
scrolls · 16 cases total
Aggregatessuite of 16 cases · mean latency
HardwareBest knownImplementationRuns
NVIDIA B200 · env 1238.7 µsRIAC as well67NVIDIA B200 · env 2257.6 µsCoral@HumanAgentSociety61

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 · 67 results · last observed 2026-08-24Record history →
#
Implementation
Latency
vs #1
Trust
Observed
1
RIAC as well
238.7µs79% SOL
1.00×
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 →
2
Amir M. Mir | SF Tensor
262.8µs77% SOL
1.10×
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 →
3
TH3CHARLie | TileGym
266.3µs77% SOL
1.12×
Reported · license unknown · no source
2026-07-21

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

no sourcelicense unknownno install recipeRun detail →
4
Isaac Sun (AES)
267.3µs77% SOL
1.12×
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 →
5
xuanzhengdu
269.7µs77% SOL
1.13×
Reported · license unknown · no source
2026-08-09

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

no sourcelicense unknownno install recipeRun detail →
6
xuanzhengdu
269.8µs77% SOL
1.13×
Reported · license unknown · no source
2026-08-09

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

no sourcelicense unknownno install recipeRun detail →
7
Databricks
270.9µs77% 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 →
8
xuanzhengdu
271.3µs77% SOL
1.14×
Reported · license unknown · no source
2026-08-09

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

no sourcelicense unknownno install recipeRun detail →
9
xuanzhengdu
271.4µs77% SOL
1.14×
Reported · license unknown · no source
2026-08-09

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

no sourcelicense unknownno install recipeRun detail →
10
xuanzhengdu
271.9µs77% SOL
1.14×
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 →
11
xuanzhengdu
272.0µs77% SOL
1.14×
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 →
12
xuanzhengdu
272.7µs77% SOL
1.14×
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 →
13
xuanzhengdu
272.8µs77% SOL
1.14×
Reported · license unknown · no source
2026-08-09

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

no sourcelicense unknownno install recipeRun detail →
14
multiagent
273.4µs77% SOL
1.15×
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 →
15
multiagent
273.5µs77% SOL
1.15×
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 →
16
TH3CHARLie | TileGym
278.0µs75% SOL
1.16×
Reported · license unknown · no source
2026-07-21

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

no sourcelicense unknownno install recipeRun detail →
17
Hyra
278.2µs76% SOL
1.17×
Reported · license unknown · no source
2026-07-14

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

no sourcelicense unknownno install recipeRun detail →
18
xuanzhengdu
279.0µs76% SOL
1.17×
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 →
19
xuanzhengdu
280.3µs76% SOL
1.17×
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 →
20
multiagent
280.8µs76% SOL
1.18×
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 →
21
xuanzhengdu
284.0µs76% SOL
1.19×
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 →
22
Isaac S
287.4µs75% SOL
1.20×
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 →
23
xuanzhengdu
292.1µs75% SOL
1.22×
Reported · license unknown · no source
2026-08-02

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

no sourcelicense unknownno install recipeRun detail →
24
doubleAI
295.9µs75% SOL
1.24×
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 →
25
xuanzhengdu
299.8µs74% SOL
1.26×
Reported · license unknown · no source
2026-08-01

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

no sourcelicense unknownno install recipeRun detail →
26
Amir M. Mir | SF Tensor
307.9µs74% SOL
1.29×
Reported · license unknown · no source
2026-07-13

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

no sourcelicense unknownno install recipeRun detail →
27
xuanzhengdu
308.0µs74% SOL
1.29×
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 →
28
xuanzhengdu
317.9µs73% SOL
1.33×
Reported · license unknown · no source
2026-08-01

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

no sourcelicense unknownno install recipeRun detail →
29
ac4k
329.0µs72% SOL
1.38×
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 →
30
Recursive
333.1µs72% SOL
1.40×
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 →
31
Amir M. Mir | SF Tensor
363.1µs69% SOL
1.52×
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 →
32
Geometric
379.7µs68% SOL
1.59×
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 →
33
Geometric
380.6µs68% SOL
1.59×
Reported · license unknown · no source
2026-08-23

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

no sourcelicense unknownno install recipeRun detail →
34
Geometric
387.2µs67% SOL
1.62×
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 →
35
Geometric
388.0µs67% SOL
1.63×
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 →
36
Geometric
388.1µs67% SOL
1.63×
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 →
37
Geometric
389.2µs67% SOL
1.63×
Reported · license unknown · no source
2026-08-19

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

no sourcelicense unknownno install recipeRun detail →
38
Geometric
389.7µs67% SOL
1.63×
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 →
39
Geometric
390.1µs67% SOL
1.63×
Reported · license unknown · no source
2026-08-20

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

no sourcelicense unknownno install recipeRun detail →
40
Geometric
390.4µs67% SOL
1.64×
Reported · license unknown · no source
2026-08-20

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

no sourcelicense unknownno install recipeRun detail →
41
Geometric
390.4µs67% SOL
1.64×
Reported · license unknown · no source
2026-08-24

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

no sourcelicense unknownno install recipeRun detail →
42
Geometric
390.8µs67% SOL
1.64×
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 →
43
Geometric
394.6µs67% SOL
1.65×
Reported · license unknown · no source
2026-08-19

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

no sourcelicense unknownno install recipeRun detail →
44
Geometric
394.9µs67% SOL
1.65×
Reported · license unknown · no source
2026-08-16

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

no sourcelicense unknownno install recipeRun detail →
45
Geometric
395.7µs67% SOL
1.66×
Reported · license unknown · no source
2026-08-24

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

no sourcelicense unknownno install recipeRun detail →
46
Geometric
398.1µs67% SOL
1.67×
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 →
47
Geometric
398.7µs67% SOL
1.67×
Reported · license unknown · no source
2026-08-19

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

no sourcelicense unknownno install recipeRun detail →
48
Geometric
404.2µs66% SOL
1.69×
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 →
49
Geometric
408.5µs66% SOL
1.71×
Reported · license unknown · no source
2026-08-13

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

no sourcelicense unknownno install recipeRun detail →
50
Geometric
429.3µs65% SOL
1.80×
Reported · license unknown · no source
2026-08-13

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

no sourcelicense unknownno install recipeRun detail →
51
Samuel Mausberg
438.4µs64% SOL
1.84×
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 →
52
Geometric
440.0µs64% SOL
1.84×
Reported · license unknown · no source
2026-08-08

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

no sourcelicense unknownno install recipeRun detail →
53
Noble Panther
440.1µs64% SOL
1.84×
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 →
54
ling0322
441.2µs64% SOL
1.85×
Reported · license unknown · no source
2026-08-05

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

no sourcelicense unknownno install recipeRun detail →
55
xuanzhengdu
446.3µs64% SOL
1.87×
Reported · license unknown · no source
2026-07-29

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

no sourcelicense unknownno install recipeRun detail →
56
Geometric
452.9µs63% SOL
1.90×
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 →
57
Runic Cheetah
471.1µs62% SOL
1.97×
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 →
58
Samuel Mausberg
478.6µs61% SOL
2.00×
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 →
59
Geometric
480.9µs61% SOL
2.01×
Reported · license unknown · no source
2026-08-05

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

no sourcelicense unknownno install recipeRun detail →
60
Runic Cheetah
484.3µs61% SOL
2.03×
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 →

7 more rows in this comparison group. Open all in search →

Implementations

Implementation
Runtime
Best latency
Evidence
Availability
Coral@HumanAgentSociety
unknown
257.6µs
1.08×
Reported
license unknown · no source
Databricks
unknown
263.6µs
1.10×
Reported
license unknown · no source
Recursive
unknown
268.0µs
1.12×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
262.8µs
1.10×
Reported
license unknown · no source
TH3CHARLie | TileGym
unknown
266.3µs
1.12×
Reported
license unknown · no source
multiagent
unknown
273.4µs
1.15×
Reported
license unknown · no source
Hyra
unknown
278.2µs
1.17×
Reported
license unknown · no source
doubleAI
unknown
289.5µs
1.21×
Reported
license unknown · no source
Show all 128 implementations ›
Isaac S
unknown
287.4µs
1.20×
Reported
license unknown · no source
ac4k
unknown
319.2µs
1.34×
Reported
license unknown · no source
Stellar Flamingo
unknown
333.0µs
1.40×
Reported
license unknown · no source
Rogue Pegasus
unknown
363.9µs
1.52×
Reported
license unknown · no source
Igor Bragaia
unknown
376.2µs
1.58×
Reported
license unknown · no source
AKO4ALL
unknown
376.5µs
1.58×
Reported
license unknown · no source
Argus
unknown
391.0µs
1.64×
Reported
license unknown · no source
Geometric
unknown
379.7µs
1.59×
Reported
license unknown · no source
XG
unknown
401.4µs
1.68×
Reported
license unknown · no source
Zesty Toucan
unknown
386.6µs
1.62×
Reported
license unknown · no source
ling0322
unknown
441.2µs
1.85×
Reported
license unknown · no source
Samuel Mausberg
unknown
438.4µs
1.84×
Reported
license unknown · no source
Noble Panther
unknown
440.1µs
1.84×
Reported
license unknown · no source
Runic Cheetah
unknown
471.1µs
1.97×
Reported
license unknown · no source
Phantom Orca
unknown
504.0µs
2.11×
Reported
license unknown · no source
Shawn_Ming
unknown
523.5µs
2.19×
Reported
license unknown · no source
wychi888
unknown
524.2µs
2.20×
Reported
license unknown · no source
Gilded Quail
unknown
522.8µs
2.19×
Reported
license unknown · no source
Bob Huang
unknown
525.9µs
2.20×
Reported
license unknown · no source
Rapid Lynx
unknown
722.8µs
3.03×
Reported
license unknown · no source
Cursor
unknown
734.2µs
3.08×
Reported
license unknown · no source
Ubios-Agent
unknown
752.3µs
3.15×
Reported
license unknown · no source
Quick Impala
unknown
1.01ms
4.21×
Reported
license unknown · no source
Clever Scorpion
unknown
1.30ms
5.46×
Reported
license unknown · no source
Zerg AI
unknown
2.24ms
9.37×
Reported
license unknown · no source
Yuang
unknown
2.77ms
11.59×
Reported
license unknown · no source
Theoric.com
unknown
4.54ms
19.04×
Reported
license unknown · no source
Databricks
unknown
270.9µs
1.13×
Reported
license unknown · no source
xuanzhengdu
unknown
269.8µs
1.13×
Reported
license unknown · no source
xuanzhengdu
unknown
271.3µs
1.14×
Reported
license unknown · no source
xuanzhengdu
unknown
271.4µs
1.14×
Reported
license unknown · no source
xuanzhengdu
unknown
271.9µs
1.14×
Reported
license unknown · no source
xuanzhengdu
unknown
272.0µs
1.14×
Reported
license unknown · no source
xuanzhengdu
unknown
272.7µs
1.14×
Reported
license unknown · no source
xuanzhengdu
unknown
272.8µs
1.14×
Reported
license unknown · no source
multiagent
unknown
273.5µs
1.15×
Reported
license unknown · no source
xuanzhengdu
unknown
279.0µs
1.17×
Reported
license unknown · no source
xuanzhengdu
unknown
280.3µs
1.17×
Reported
license unknown · no source
multiagent
unknown
280.8µs
1.18×
Reported
license unknown · no source
xuanzhengdu
unknown
284.0µs
1.19×
Reported
license unknown · no source
TH3CHARLie | TileGym
unknown
278.0µs
1.16×
Reported
license unknown · no source
xuanzhengdu
unknown
292.1µs
1.22×
Reported
license unknown · no source
doubleAI
unknown
295.9µs
1.24×
Reported
license unknown · no source
xuanzhengdu
unknown
299.8µs
1.26×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
307.9µs
1.29×
Reported
license unknown · no source
xuanzhengdu
unknown
308.0µs
1.29×
Reported
license unknown · no source
xuanzhengdu
unknown
317.9µs
1.33×
Reported
license unknown · no source
ac4k
unknown
329.0µs
1.38×
Reported
license unknown · no source
doubleAI
unknown
324.2µs
1.36×
Reported
license unknown · no source
Recursive
unknown
333.1µs
1.40×
Reported
license unknown · no source
Stellar Flamingo
unknown
347.3µs
1.45×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
353.3µs
1.48×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
363.1µs
1.52×
Reported
license unknown · no source
Igor Bragaia
unknown
376.4µs
1.58×
Reported
license unknown · no source
Igor Bragaia
unknown
381.8µs
1.60×
Reported
license unknown · no source
Geometric
unknown
380.6µs
1.59×
Reported
license unknown · no source
Geometric
unknown
387.2µs
1.62×
Reported
license unknown · no source
Geometric
unknown
388.0µs
1.63×
Reported
license unknown · no source
Geometric
unknown
388.1µs
1.63×
Reported
license unknown · no source
Geometric
unknown
389.2µs
1.63×
Reported
license unknown · no source
Geometric
unknown
389.7µs
1.63×
Reported
license unknown · no source
Geometric
unknown
390.1µs
1.63×
Reported
license unknown · no source
Geometric
unknown
390.4µs
1.64×
Reported
license unknown · no source
Geometric
unknown
390.4µs
1.64×
Reported
license unknown · no source
Geometric
unknown
390.8µs
1.64×
Reported
license unknown · no source
Geometric
unknown
394.6µs
1.65×
Reported
license unknown · no source
Geometric
unknown
395.7µs
1.66×
Reported
license unknown · no source
Geometric
unknown
394.9µs
1.65×
Reported
license unknown · no source
Geometric
unknown
398.1µs
1.67×
Reported
license unknown · no source
Geometric
unknown
398.7µs
1.67×
Reported
license unknown · no source
Geometric
unknown
404.2µs
1.69×
Reported
license unknown · no source
Geometric
unknown
408.5µs
1.71×
Reported
license unknown · no source
Stellar Flamingo
unknown
436.7µs
1.83×
Reported
license unknown · no source
XG
unknown
441.2µs
1.85×
Reported
license unknown · no source
Geometric
unknown
429.3µs
1.80×
Reported
license unknown · no source
Igor Bragaia
unknown
443.9µs
1.86×
Reported
license unknown · no source
Igor Bragaia
unknown
435.4µs
1.82×
Reported
license unknown · no source
Geometric
unknown
440.0µs
1.84×
Reported
license unknown · no source
xuanzhengdu
unknown
446.3µs
1.87×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
463.7µs
1.94×
Reported
license unknown · no source
Geometric
unknown
452.9µs
1.90×
Reported
license unknown · no source
XG
unknown
481.8µs
2.02×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
485.7µs
2.03×
Reported
license unknown · no source
Igor Bragaia
unknown
477.6µs
2.00×
Reported
license unknown · no source
AKO4ALL
unknown
485.3µs
2.03×
Reported
license unknown · no source
Geometric
unknown
480.9µs
2.01×
Reported
license unknown · no source
Runic Cheetah
unknown
484.3µs
2.03×
Reported
license unknown · no source
Samuel Mausberg
unknown
478.6µs
2.00×
Reported
license unknown · no source
Stellar Flamingo
unknown
496.7µs
2.08×
Reported
license unknown · no source
Geometric
unknown
503.6µs
2.11×
Reported
license unknown · no source
Gilded Quail
unknown
525.5µs
2.20×
Reported
license unknown · no source
Samuel Mausberg
unknown
538.2µs
2.25×
Reported
license unknown · no source
AKO4ALL
unknown
599.5µs
2.51×
Reported
license unknown · no source
Rogue Pegasus
unknown
643.0µs
2.69×
Reported
license unknown · no source
Bob Huang
unknown
658.6µs
2.76×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
721.0µs
3.02×
Reported
license unknown · no source
Geometric
unknown
716.6µs
3.00×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
748.4µs
3.13×
Reported
license unknown · no source
Cursor
unknown
759.3µs
3.18×
Reported
license unknown · no source
Geometric
unknown
780.0µs
3.27×
Reported
license unknown · no source
Igor Bragaia
unknown
792.0µs
3.32×
Reported
license unknown · no source
Igor Bragaia
unknown
812.3µs
3.40×
Reported
license unknown · no source
Igor Bragaia
unknown
863.4µs
3.62×
Reported
license unknown · no source
Igor Bragaia
unknown
1.02ms
4.27×
Reported
license unknown · no source
Igor Bragaia
unknown
1.07ms
4.50×
Reported
license unknown · no source
Rapid Lynx
unknown
1.07ms
4.49×
Reported
license unknown · no source
Rapid Lynx
unknown
1.12ms
4.67×
Reported
license unknown · no source
Igor Bragaia
unknown
1.25ms
5.25×
Reported
license unknown · no source
Rapid Lynx
unknown
1.30ms
5.45×
Reported
license unknown · no source
Rapid Lynx
unknown
1.32ms
5.55×
Reported
license unknown · no source
Igor Bragaia
unknown
1.44ms
6.05×
Reported
license unknown · no source
Igor Bragaia
unknown
1.44ms
6.05×
Reported
license unknown · no source
Quick Impala
unknown
2.38ms
9.95×
Reported
license unknown · no source
Bob Huang
unknown
3.85ms
16.14×
Reported
license unknown · no source
Yuang
unknown
3.97ms
16.61×
Reported
license unknown · no source
Coral@HumanAgentSociety
unknown
16.0ms
67.21×
Reported
license unknown · no source
Geometric
unknown
42.7ms
178.75×
Reported
license unknown · no source
RIAC as well
unknown
238.7µs
1.00×
Reported
license unknown · no source
Isaac Sun (AES)
unknown
267.3µs
1.12×
Reported
license unknown · no source
xuanzhengdu
unknown
269.7µs
1.13×
Reported
license unknown · no source

Semantics

Inputs and outputs
grad_attn_outputbf16 [batch_size, seq_len_q, num_attention_heads, head_dim]
attn_weightsbf16 [batch_size, num_attention_heads, seq_len_q, seq_len_kv]
attn_weights_droppedbf16 [batch_size, num_attention_heads, seq_len_q, seq_len_kv]
value_statesbf16 [batch_size, num_key_value_heads, seq_len_kv, head_dim]
dropout_maskbool [batch_size, num_attention_heads, seq_len_q, seq_len_kv]
attention_dropoutfp32 scalar
grad_attn_scoresbf16 [batch_size, num_attention_heads, seq_len_q, seq_len_kv]
grad_value_statesbf16 [batch_size, num_key_value_heads, seq_len_kv, head_dim]
Axes and behavior
head_dimconstant = 128
seq_len_qvariable
batch_sizevariable
seq_len_kvvariable
num_attention_headsconstant = 80
num_key_value_headsconstant = 8
num_key_value_groupsnum_key_value_groups = num_attention_heads // num_key_value_heads
determinismunspecified
constraintsNo mutation or aliasing
Identity
alias001_attention_softmax_dropout_value_matmul_backwardmodelseed-seed-oss-36b-instructsha2561201a1ddd669…
Sources: NVIDIA SOL-ExecBench (2026-08-24)last observed 2026-08-24How records are decidedJSON