Skip to content
KernelIndex
Search⌘K
Sign in

SAM hq mask decoder IoU hypernetwork fusion

37 eligible runs
vision

SAM-HQ's fused IoU prediction and hypernetwork weight generation from transformer output tokens. Processes transformer outputs to simultaneously generate IoU quality scores via a 3-layer MLP (256->1024->1024->4) and dynamic convolution weights for mask generation via 4 parallel 3-layer MLPs (256->256->256->32).

Fastest reported
9.42µsmean · 94% of SOL
RIAC as wellLicense unknown · unknown

Reported evidence · last observed 2026-08-20. Reported by source; not independently reproduced. No public source. License unknown.

No public sourceRun detail →

Current records

Workloadsuite of 16 cases · mean latencyhidden_size = 256 · num_mask_tokens = 4 · output_channels = 32 · iou_head_hidden_dim = 102416 cases
point_batch_size
batch_size
dtype
11fp3214fp32132fp3222fp3224fp32264fp3231fp3232fp3234fp3244fp3248fp32416fp3251fp3271fp3282fp32101fp32
scrolls · 16 cases total
Aggregatessuite of 16 cases · mean latency
HardwareBest knownImplementationRuns
NVIDIA B200 · env 19.42 µsRIAC as well27NVIDIA B200 · env 215.5 µsDatabricks10

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 · 27 results · last observed 2026-08-20Record history →
#
Implementation
Latency
vs #1
Trust
Observed
1
RIAC as well
9.42µs94% SOL
1.00×
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 →
2
Amir M. Mir | SF Tensor
10.9µs93% SOL
1.15×
Reported · license unknown · no source
2026-07-26

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

no sourcelicense unknownno install recipeRun detail →
3
Databricks
15.8µs91% SOL
1.68×
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
Geometric
16.2µs91% SOL
1.72×
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 →
5
Recursive
18.0µs89% SOL
1.91×
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
Hyra
19.5µs89% SOL
2.08×
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 →
7
doubleAI
20.2µs88% SOL
2.15×
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
Geometric
20.8µs88% 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 →
9
ac4k
20.9µs88% SOL
2.22×
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 →
10
Geometric
23.5µs87% SOL
2.50×
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 →
11
Geometric
23.6µs87% SOL
2.51×
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 →
12
Amir M. Mir | SF Tensor
23.8µs86% SOL
2.52×
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 →
13
ac4k
23.9µs87% SOL
2.54×
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 →
14
Geometric
25.8µs86% SOL
2.74×
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 →
15
Geometric
26.1µs85% SOL
2.77×
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 →
16
TH3CHARLie | TileGym
29.2µs84% SOL
3.10×
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 →
17
Geometric
39.4µs79% SOL
4.18×
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 →
18
Qin Liu @ UCD
40.6µs79% SOL
4.31×
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 →
19
Geometric
61.3µs71% SOL
6.50×
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 →
20
Qin Liu @ UCD
69.3µs69% SOL
7.36×
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 →
21
Geometric
72.1µs68% SOL
7.66×
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 →
22
Geometric
75.1µs66% SOL
7.98×
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 →
23
Qin Liu @ UCD
81.4µs65% SOL
8.64×
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 →
24
Qin Liu @ UCD
102.0µs60% SOL
10.83×
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 →
25
Qin Liu @ UCD
134.9µs53% SOL
14.33×
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 →
26
Cursor
197.5µs43% SOL
20.96×
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 →
27
Qin Liu @ UCD
205.2µs43% SOL
21.79×
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 →

Implementations

Implementation
Runtime
Best latency
Evidence
Availability
RIAC as well
unknown
9.42µs
1.00×
Reported
license unknown · no source
Databricks
unknown
15.5µs
1.65×
Reported
license unknown · no source
Geometric
unknown
16.2µs
1.72×
Reported
license unknown · no source
Recursive
unknown
17.5µs
1.86×
Reported
license unknown · no source
doubleAI
unknown
19.7µs
2.09×
Reported
license unknown · no source
ac4k
unknown
20.9µs
2.22×
Reported
license unknown · no source
TH3CHARLie | TileGym
unknown
29.2µs
3.10×
Reported
license unknown · no source
Qin Liu @ UCD
unknown
40.6µs
4.31×
Reported
license unknown · no source
Show all 37 implementations ›
Argus
unknown
177.4µs
18.83×
Reported
license unknown · no source
Cursor
unknown
193.4µs
20.54×
Reported
license unknown · no source
Databricks
unknown
15.8µs
1.68×
Reported
license unknown · no source
Recursive
unknown
18.0µs
1.91×
Reported
license unknown · no source
doubleAI
unknown
20.2µs
2.15×
Reported
license unknown · no source
Geometric
unknown
20.8µs
2.21×
Reported
license unknown · no source
ac4k
unknown
22.3µs
2.37×
Reported
license unknown · no source
doubleAI
unknown
23.1µs
2.46×
Reported
license unknown · no source
Geometric
unknown
23.6µs
2.51×
Reported
license unknown · no source
ac4k
unknown
23.9µs
2.54×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
23.8µs
2.52×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
23.9µs
2.54×
Reported
license unknown · no source
Geometric
unknown
25.8µs
2.74×
Reported
license unknown · no source
Geometric
unknown
26.1µs
2.77×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
28.9µs
3.07×
Reported
license unknown · no source
Geometric
unknown
39.4µs
4.18×
Reported
license unknown · no source
Geometric
unknown
61.3µs
6.50×
Reported
license unknown · no source
Qin Liu @ UCD
unknown
69.3µs
7.36×
Reported
license unknown · no source
Geometric
unknown
72.1µs
7.66×
Reported
license unknown · no source
Geometric
unknown
75.1µs
7.98×
Reported
license unknown · no source
Qin Liu @ UCD
unknown
81.4µs
8.64×
Reported
license unknown · no source
Qin Liu @ UCD
unknown
102.0µs
10.83×
Reported
license unknown · no source
Qin Liu @ UCD
unknown
134.9µs
14.33×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
192.5µs
20.44×
Reported
license unknown · no source
Cursor
unknown
197.5µs
20.96×
Reported
license unknown · no source
Qin Liu @ UCD
unknown
205.2µs
21.79×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
10.9µs
1.15×
Reported
license unknown · no source
Hyra
unknown
19.5µs
2.08×
Reported
license unknown · no source
Geometric
unknown
23.5µs
2.50×
Reported
license unknown · no source

Semantics

Inputs and outputs
iou_token_outfp32 [batch_size, point_batch_size, hidden_size]
mask_tokens_outfp32 [batch_size, point_batch_size, num_mask_tokens, hidden_size]
iou_proj_in_weightfp32 [iou_head_hidden_dim, hidden_size]
iou_proj_in_biasfp32 [iou_head_hidden_dim]
iou_hidden_weightfp32 [iou_head_hidden_dim, iou_head_hidden_dim]
iou_hidden_biasfp32 [iou_head_hidden_dim]
iou_proj_out_weightfp32 [num_mask_tokens, iou_head_hidden_dim]
iou_proj_out_biasfp32 [num_mask_tokens]
hyper0_proj_in_weightfp32 [hidden_size, hidden_size]
hyper0_proj_in_biasfp32 [hidden_size]
hyper0_hidden_weightfp32 [hidden_size, hidden_size]
hyper0_hidden_biasfp32 [hidden_size]
hyper0_proj_out_weightfp32 [output_channels, hidden_size]
hyper0_proj_out_biasfp32 [output_channels]
hyper1_proj_in_weightfp32 [hidden_size, hidden_size]
hyper1_proj_in_biasfp32 [hidden_size]
hyper1_hidden_weightfp32 [hidden_size, hidden_size]
hyper1_hidden_biasfp32 [hidden_size]
hyper1_proj_out_weightfp32 [output_channels, hidden_size]
hyper1_proj_out_biasfp32 [output_channels]
hyper2_proj_in_weightfp32 [hidden_size, hidden_size]
hyper2_proj_in_biasfp32 [hidden_size]
hyper2_hidden_weightfp32 [hidden_size, hidden_size]
hyper2_hidden_biasfp32 [hidden_size]
hyper2_proj_out_weightfp32 [output_channels, hidden_size]
hyper2_proj_out_biasfp32 [output_channels]
hyper3_proj_in_weightfp32 [hidden_size, hidden_size]
hyper3_proj_in_biasfp32 [hidden_size]
hyper3_hidden_weightfp32 [hidden_size, hidden_size]
hyper3_hidden_biasfp32 [hidden_size]
hyper3_proj_out_weightfp32 [output_channels, hidden_size]
hyper3_proj_out_biasfp32 [output_channels]
iou_scoresfp32 [batch_size, point_batch_size, num_mask_tokens]
hyper_weightsfp32 [batch_size, point_batch_size, num_mask_tokens, output_channels]
Axes and behavior
batch_sizevariable
hidden_sizeconstant = 256
num_mask_tokensconstant = 4
output_channelsconstant = 32
point_batch_sizevariable
iou_head_hidden_dimconstant = 1024
determinismunspecified
constraintsNo mutation or aliasing
Identity
alias086_sam_hq_mask_decoder_iou_hypernetwork_fusionmodelsam-hq-vit-basesha25622ec49b2da29…
Sources: NVIDIA SOL-ExecBench (2026-08-20)last observed 2026-08-20How records are decidedJSON