Skip to content
KernelIndex
Search⌘K
Sign in

Cross attention text video conditioning backward

34 eligible runs
video

Backward pass for cross-attention text-video conditioning. Computes gradients through output projection, attention mechanism (softmax, Q@K^T), and Q/K/V linear projections. Queries from video latents, keys/values from text embeddings.

Fastest reported
114.5µsmean · 75% of SOL
RIAC as wellLicense unknown · unknown

Reported evidence · last observed 2026-08-18. Reported by source; not independently reproduced. No public source. License unknown.

No public sourceRun detail →

Current records

Workloadsuite of 16 cases · mean latencyhead_dim = 64 · num_heads = 16 · hidden_size = 102416 cases
num_video_tokens
batch_size
num_text_tokens
dtype
1283277fp321286477fp32256277fp322562211fp322564293fp322568373fp32512177fp32512277fp32512477fp32512877fp32691277fp32853477fp321024277fp322048277fp324096177fp324096277fp32
scrolls · 16 cases total
Aggregatessuite of 16 cases · mean latency
HardwareBest knownImplementationRuns
NVIDIA B200 · env 1114.5 µsRIAC as well22NVIDIA B200 · env 2162.1 µsRecursive12

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 · 22 results · last observed 2026-08-24Record history →
#
Implementation
Latency
vs #1
Trust
Observed
1
RIAC as well
114.5µs75% SOL
1.00×
Reported · license unknown · no source
2026-08-18

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

no sourcelicense unknownno install recipeRun detail →
2
Amir M. Mir | SF Tensor
144.7µs70% SOL
1.26×
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
Hyra
150.3µs69% SOL
1.31×
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 →
4
ac4k
154.4µs68% SOL
1.35×
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 →
5
Recursive
168.7µs66% SOL
1.47×
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
doubleAI
169.4µs66% SOL
1.48×
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 →
7
Proud Wren @ NVIDIA
178.5µs65% SOL
1.56×
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
182.0µs64% SOL
1.59×
Reported · license unknown · no source
2026-08-18

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

no sourcelicense unknownno install recipeRun detail →
9
Geometric
182.2µs64% SOL
1.59×
Reported · license unknown · no source
2026-08-18

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

no sourcelicense unknownno install recipeRun detail →
10
Geometric
182.4µs64% SOL
1.59×
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 →
11
Geometric
183.4µs64% SOL
1.60×
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 →
12
Geometric
209.8µs61% SOL
1.83×
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 →
13
Geometric
214.4µs60% SOL
1.87×
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 →
14
Amir M. Mir | SF Tensor
216.4µs60% SOL
1.89×
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 →
15
Amir M. Mir | SF Tensor
235.0µs58% SOL
2.05×
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 →
16
Geometric
242.9µs57% SOL
2.12×
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 →
17
Geometric
263.9µs55% SOL
2.30×
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 →
18
Amir M. Mir | SF Tensor
296.2µs52% 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 →
19
Cursor
331.5µs49% SOL
2.89×
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 →
20
Geometric
384.9µs46% SOL
3.36×
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 →
21
Geometric
408.2µs44% SOL
3.56×
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 →
22
Geometric
2.59ms11% SOL
22.65×
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 →

Implementations

Implementation
Runtime
Best latency
Evidence
Availability
RIAC as well
unknown
114.5µs
1.00×
Reported
license unknown · no source
Recursive
unknown
162.1µs
1.42×
Reported
license unknown · no source
doubleAI
unknown
164.3µs
1.43×
Reported
license unknown · no source
Proud Wren @ NVIDIA
unknown
178.0µs
1.55×
Reported
license unknown · no source
Geometric
unknown
182.0µs
1.59×
Reported
license unknown · no source
Cursor
unknown
318.2µs
2.78×
Reported
license unknown · no source
bismuth83
unknown
552.0µs
4.82×
Reported
license unknown · no source
Recursive
unknown
168.7µs
1.47×
Reported
license unknown · no source
Show all 34 implementations ›
doubleAI
unknown
169.4µs
1.48×
Reported
license unknown · no source
Proud Wren @ NVIDIA
unknown
178.5µs
1.56×
Reported
license unknown · no source
Geometric
unknown
182.2µs
1.59×
Reported
license unknown · no source
Geometric
unknown
182.4µs
1.59×
Reported
license unknown · no source
Geometric
unknown
183.4µs
1.60×
Reported
license unknown · no source
Geometric
unknown
209.8µs
1.83×
Reported
license unknown · no source
Geometric
unknown
214.4µs
1.87×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
216.4µs
1.89×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
235.0µs
2.05×
Reported
license unknown · no source
Geometric
unknown
242.9µs
2.12×
Reported
license unknown · no source
Geometric
unknown
263.9µs
2.30×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
286.8µs
2.50×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
296.2µs
2.59×
Reported
license unknown · no source
Cursor
unknown
331.5µs
2.89×
Reported
license unknown · no source
Geometric
unknown
384.9µs
3.36×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
398.8µs
3.48×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
402.3µs
3.51×
Reported
license unknown · no source
Geometric
unknown
408.2µs
3.56×
Reported
license unknown · no source
doubleAI
unknown
441.0µs
3.85×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
550.3µs
4.80×
Reported
license unknown · no source
bismuth83
unknown
553.1µs
4.83×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
1.36ms
11.91×
Reported
license unknown · no source
Geometric
unknown
2.59ms
22.65×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
144.7µs
1.26×
Reported
license unknown · no source
Hyra
unknown
150.3µs
1.31×
Reported
license unknown · no source
ac4k
unknown
154.4µs
1.35×
Reported
license unknown · no source

Semantics

Inputs and outputs
grad_outputfp32 [batch_size, num_video_tokens, hidden_size]
video_latentsfp32 [batch_size, num_video_tokens, hidden_size]
text_embeddingsfp32 [batch_size, num_text_tokens, hidden_size]
query_weightfp32 [hidden_size, hidden_size]
query_biasfp32 [hidden_size]
key_weightfp32 [hidden_size, hidden_size]
key_biasfp32 [hidden_size]
value_weightfp32 [hidden_size, hidden_size]
value_biasfp32 [hidden_size]
output_weightfp32 [hidden_size, hidden_size]
output_biasfp32 [hidden_size]
scalefp32 scalar
grad_video_latentsfp32 [batch_size, num_video_tokens, hidden_size]
grad_text_embeddingsfp32 [batch_size, num_text_tokens, hidden_size]
grad_query_weightfp32 [hidden_size, hidden_size]
grad_query_biasfp32 [hidden_size]
grad_key_weightfp32 [hidden_size, hidden_size]
grad_key_biasfp32 [hidden_size]
grad_value_weightfp32 [hidden_size, hidden_size]
grad_value_biasfp32 [hidden_size]
grad_output_weightfp32 [hidden_size, hidden_size]
grad_output_biasfp32 [hidden_size]
Axes and behavior
head_dimconstant = 64
num_headsconstant = 16
batch_sizevariable
hidden_sizeconstant = 1024
num_text_tokensvariable
num_video_tokensvariable
determinismunspecified
constraintsNo mutation or aliasing
Identity
alias021_cross_attention_text_video_conditioning_backwardmodelwan2-2-t2v-a14bsha256904c37c738c9…
Sources: NVIDIA SOL-ExecBench (2026-08-24)last observed 2026-08-24How records are decidedJSON