Skip to content
KernelIndex
Search⌘K

Altup predict correction cycle backward

29 eligible runs
other

Backward pass for AltUp predict-correct cycle. Computes gradients for prediction coefficients, correction coefficients, router weights, and norm weights through the combined predict and correct operations.

Fastest reported
56.4µsmean · 97% of SOL
Amir M. Mir | SF TensorLicense unknown · unknown

Reported evidence · last observed 2026-08-19. Reported by source; not independently reproduced. No public source. License unknown.

No public sourceRun detail →

Current records

Workloadsuite of 16 cases · mean latencyhidden_size = 2304 · altup_num_inputs = 3 · altup_num_inputs_sq = 916 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 · 20 results · last observed 2026-08-24Record history →
#
Implementation
Latency
vs #1
Trust
Observed
1
56.4µs97% SOL
1.00×
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 →
2
71.3µs95% SOL
1.26×
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 →
3
78.9µs94% SOL
1.40×
Reported · license unknown · no source
2026-07-12

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

no sourcelicense unknownno install recipeRun detail →
4
86.8µs93% SOL
1.54×
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 →
5
91.3µs92% SOL
1.62×
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
92.8µs92% SOL
1.65×
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 →
7
100.5µs91% SOL
1.78×
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 →
8
109.6µs90% SOL
1.94×
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
111.7µs90% SOL
1.98×
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 →
10
118.1µs89% SOL
2.09×
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 →
11
122.1µs89% SOL
2.16×
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 →
12
130.0µs88% SOL
2.30×
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 →
13
135.9µs87% SOL
2.41×
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 →
14
164.4µs84% SOL
2.91×
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 →
15
184.4µs82% SOL
3.27×
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 →
16
254.2µs76% SOL
4.50×
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 →
17
290.4µs73% SOL
5.15×
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 →
18
528.4µs58% SOL
9.36×
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 →
19
815.8µs47% SOL
14.46×
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
3.63ms17% SOL
64.31×
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 →

Implementations

Implementation
Runtime
Best latency
Evidence
Availability
unknown
56.4µs
1.00×
Reported
license unknown · no source
unknown
88.8µs
1.57×
Reported
license unknown · no source
unknown
92.8µs
1.65×
Reported
license unknown · no source
unknown
109.6µs
1.94×
Reported
license unknown · no source
unknown
183.7µs
3.26×
Reported
license unknown · no source
unknown
729.1µs
12.92×
Reported
license unknown · no source
unknown
86.8µs
1.54×
Reported
license unknown · no source
unknown
91.3µs
1.62×
Reported
license unknown · no source
Show all 29 implementations ›
unknown
100.5µs
1.78×
Reported
license unknown · no source
unknown
111.7µs
1.98×
Reported
license unknown · no source
unknown
118.1µs
2.09×
Reported
license unknown · no source
unknown
122.1µs
2.16×
Reported
license unknown · no source
unknown
130.0µs
2.30×
Reported
license unknown · no source
unknown
135.9µs
2.41×
Reported
license unknown · no source
unknown
164.4µs
2.91×
Reported
license unknown · no source
unknown
184.4µs
3.27×
Reported
license unknown · no source
unknown
181.4µs
3.21×
Reported
license unknown · no source
unknown
254.2µs
4.50×
Reported
license unknown · no source
unknown
289.4µs
5.13×
Reported
license unknown · no source
unknown
290.4µs
5.15×
Reported
license unknown · no source
unknown
528.4µs
9.36×
Reported
license unknown · no source
unknown
611.5µs
10.84×
Reported
license unknown · no source
unknown
731.2µs
12.96×
Reported
license unknown · no source
unknown
784.7µs
13.91×
Reported
license unknown · no source
unknown
815.8µs
14.46×
Reported
license unknown · no source
unknown
3.63ms
64.31×
Reported
license unknown · no source
unknown
71.3µs
1.26×
Reported
license unknown · no source
unknown
78.6µs
1.39×
Reported
license unknown · no source
unknown
78.9µs
1.40×
Reported
license unknown · no source

Semantics

Inputs and outputs
grad_correctedbf16 [altup_num_inputs, batch_size, seq_len, hidden_size]
hidden_statesbf16 [altup_num_inputs, batch_size, seq_len, hidden_size]
activatedbf16 [batch_size, seq_len, hidden_size]
prediction_coef_weightbf16 [altup_num_inputs_sq, altup_num_inputs]
correction_coef_weightbf16 [altup_num_inputs, altup_num_inputs]
router_weightbf16 [altup_num_inputs, hidden_size]
norm_weightbf16 [hidden_size]
altup_active_idxint32 scalar
rms_norm_epsfp32 scalar
grad_hidden_statesbf16 [altup_num_inputs, batch_size, seq_len, hidden_size]
grad_activatedbf16 [batch_size, seq_len, hidden_size]
grad_prediction_coef_weightfp32 [altup_num_inputs_sq, altup_num_inputs]
grad_correction_coef_weightfp32 [altup_num_inputs, altup_num_inputs]
grad_router_weightfp32 [altup_num_inputs, hidden_size]
grad_norm_weightfp32 [hidden_size]
Axes and behavior
seq_lenvariable
batch_sizevariable
hidden_sizeconstant = 2304
altup_num_inputsconstant = 3
altup_num_inputs_sqaltup_num_inputs_sq = altup_num_inputs * altup_num_inputs
determinismunspecified
constraintsNo mutation or aliasing
Identity
alias040_altup_predict_correction_cycle_backwardmodelgemma-3n-e4bsha256ab84ef40da55…
Sources: NVIDIA SOL-ExecBench (2026-08-24)last observed 2026-08-24How records are decidedJSON