VAE decoder mid block attention ResNet

28 eligible runs

diffusion

VAE decoder mid-block combining ResNet blocks with self-attention at the bottleneck. Processes compressed latent representation through: ResNet -> Attention -> ResNet. The attention operates on spatial features with GroupNorm, QKV projections, multi-head self-attention, and residual connections. Uses 512 channels at the bottleneck with 4 groups for GroupNorm.

Fastest reported

271.5µsmean · 75% of SOL

RIAC as wellLicense unknown · unknown

Reported evidence · last observed 2026-08-24. Reported by source; not independently reproduced. No public source. License unknown.

No public sourceRun detail →

Current records

Workloadsuite of 11 cases · mean latencynum_heads = 512 · num_groups = 32 · in_channels = 512 · temb_channels = 512 · attention_head_dim = 111 cases

batch_size

height

width

dtype

11616fp32 13232fp32 14848fp32 16161fp32 24141fp32 26464fp32 41616fp32 44848fp32 83232fp32 163232fp32 323232fp32

Aggregatessuite of 11 cases · mean latency

HardwareBest knownImplementationRuns

NVIDIA B200 · env 1271.5 µsRIAC as well21 NVIDIA B200 · env 2349.9 µsdoubleAI7

Not measured on H100 for this workload. Challenges →

Source-native comparison · GPU NVIDIA B200 · Workload suite of 11 cases · mean latency · Protocol SOL-ExecBench evaluation stack v1.1 · mean · 21 results · last observed 2026-08-24Record history →

#

Implementation

Latency

vs #1

Trust

Observed

1

271.5µs75% SOL

1.00×

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 →

2

Amir M. Mir | SF Tensor

306.8µs73% SOL

1.13×

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

368.2µs68% SOL

1.36×

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

Proud Wren @ NVIDIA

383.2µs67% SOL

1.41×

Reported · license unknown · no source

2026-07-11

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

no sourcelicense unknownno install recipeRun detail →

5

407.4µs65% SOL

1.50×

Reported · license unknown · no source

2026-07-20

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

no sourcelicense unknownno install recipeRun detail →

6

448.6µs62% 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 →

7

458.7µs62% SOL

1.69×

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 →

8

528.4µs58% SOL

1.95×

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 →

9

566.4µs56% SOL

2.09×

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 →

10

616.4µs54% SOL

2.27×

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 →

11

650.5µs53% SOL

2.40×

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 →

12

760.0µs48% SOL

2.80×

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 →

13

771.6µs47% SOL

2.84×

Reported · license unknown · no source

2026-08-10

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

no sourcelicense unknownno install recipeRun detail →

14

785.0µs47% SOL

2.89×

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 →

15

914.7µs43% SOL

3.37×

Reported · license unknown · no source

2026-07-11

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

no sourcelicense unknownno install recipeRun detail →

16

985.5µs41% SOL

3.63×

Reported · license unknown · no source

2026-08-10

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

no sourcelicense unknownno install recipeRun detail →

17

1.54ms30% SOL

5.69×

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 →

18

1.57ms30% SOL

5.79×

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 →

19

3.21ms20% SOL

11.81×

Reported · license unknown · no source

2026-08-10

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

no sourcelicense unknownno install recipeRun detail →

20

3.29ms17% SOL

12.11×

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 →

21

28.6ms3% SOL

105.47×

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 →

Implementations

Implementation

Runtime

Best latency

Evidence

Availability

unknown

271.5µs

1.00×

Reported

license unknown · no source

unknown

366.4µs

1.35×

Reported

license unknown · no source

Proud Wren @ NVIDIA

unknown

383.2µs

1.41×

Reported

license unknown · no source

unknown

407.4µs

1.50×

Reported

license unknown · no source

unknown

448.6µs

1.65×

Reported

license unknown · no source

unknown

458.7µs

1.69×

Reported

license unknown · no source

unknown

650.5µs

2.40×

Reported

license unknown · no source

Amir M. Mir | SF Tensor

unknown

429.0µs

1.58×

Reported

license unknown · no source

Show all 28 implementations ›

unknown

528.4µs

1.95×

Reported

license unknown · no source

unknown

566.4µs

2.09×

Reported

license unknown · no source

unknown

563.1µs

2.07×

Reported

license unknown · no source

unknown

616.4µs

2.27×

Reported

license unknown · no source

unknown

768.3µs

2.83×

Reported

license unknown · no source

unknown

760.0µs

2.80×

Reported

license unknown · no source

unknown

771.6µs

2.84×

Reported

license unknown · no source

unknown

785.0µs

2.89×

Reported

license unknown · no source

unknown

914.7µs

3.37×

Reported

license unknown · no source

unknown

985.5µs

3.63×

Reported

license unknown · no source

Amir M. Mir | SF Tensor

unknown

1.17ms

4.32×

Reported

license unknown · no source

Amir M. Mir | SF Tensor

unknown

1.51ms

5.58×

Reported

license unknown · no source

unknown

1.54ms

5.69×

Reported

license unknown · no source

unknown

1.57ms

5.79×

Reported

license unknown · no source

unknown

3.21ms

11.81×

Reported

license unknown · no source

unknown

3.29ms

12.11×

Reported

license unknown · no source

unknown

28.6ms

105.47×

Reported

license unknown · no source

Amir M. Mir | SF Tensor

unknown

306.8µs

1.13×

Reported

license unknown · no source

unknown

349.9µs

1.29×

Reported

license unknown · no source

unknown

368.2µs

1.36×

Reported

license unknown · no source

Semantics

Inputs and outputs

hidden_statesfp32 [batch_size, in_channels, height, width]

tembfp32 [batch_size, temb_channels]

resnet1_norm1_weightfp32 [in_channels]

resnet1_norm1_biasfp32 [in_channels]

resnet1_conv1_weightfp32 [in_channels, in_channels, 3, 3]

resnet1_conv1_biasfp32 [in_channels]

resnet1_time_emb_proj_weightfp32 [in_channels, temb_channels]

resnet1_time_emb_proj_biasfp32 [in_channels]

resnet1_norm2_weightfp32 [in_channels]

resnet1_norm2_biasfp32 [in_channels]

resnet1_conv2_weightfp32 [in_channels, in_channels, 3, 3]

resnet1_conv2_biasfp32 [in_channels]

attn_group_norm_weightfp32 [in_channels]

attn_group_norm_biasfp32 [in_channels]

attn_to_q_weightfp32 [in_channels, in_channels]

attn_to_q_biasfp32 [in_channels]

attn_to_k_weightfp32 [in_channels, in_channels]

attn_to_k_biasfp32 [in_channels]

attn_to_v_weightfp32 [in_channels, in_channels]

attn_to_v_biasfp32 [in_channels]

attn_to_out_weightfp32 [in_channels, in_channels]

attn_to_out_biasfp32 [in_channels]

resnet2_norm1_weightfp32 [in_channels]

resnet2_norm1_biasfp32 [in_channels]

resnet2_conv1_weightfp32 [in_channels, in_channels, 3, 3]

resnet2_conv1_biasfp32 [in_channels]

resnet2_time_emb_proj_weightfp32 [in_channels, temb_channels]

resnet2_time_emb_proj_biasfp32 [in_channels]

resnet2_norm2_weightfp32 [in_channels]

resnet2_norm2_biasfp32 [in_channels]

resnet2_conv2_weightfp32 [in_channels, in_channels, 3, 3]

resnet2_conv2_biasfp32 [in_channels]

epsfp32 scalar

outputfp32 [batch_size, in_channels, height, width]

Axes and behavior

widthvariable

heightvariable

num_headsnum_heads = in_channels // attention_head_dim

batch_sizevariable

num_groupsconstant = 32

in_channelsconstant = 512

temb_channelsconstant = 512

attention_head_dimconstant = 1

determinismunspecified

constraintsNo mutation or aliasing

Identity

alias050_vae_decoder_mid_block_attention_resnetmodelflux-2-dev-vaesha256b6ec70b78dd1…

Sources: NVIDIA SOL-ExecBench (2026-08-24)last observed 2026-08-24How records are decided JSON