Skip to content
KernelIndex
Search⌘K
Sign in

Expert token scatter with weighted forward backward

29 eligible runs
moe

Backward pass for expert token scatter with weighted forward. Computes gradients for hidden_states, routing_weights, and expert MLP weights (w1, w2, w3) through SwiGLU activation.

Fastest reported
261.2µsmean · 94% of SOL
RecursiveLicense unknown · unknown

Reported evidence · last observed 2026-07-10. Reported by source; not independently reproduced. No public source. License unknown.

No public sourceRun detail →

Current records

Workloadsuite of 16 cases · mean latencyffn_dim = 14336 · hidden_dim = 409616 cases
batch_seq_len
num_tokens
dtype
25664bf1651264bf16512256bf1654164bf16997128bf161024128bf161024512bf161571196bf161721215bf162048256bf162053256bf162371296bf162689336bf163967496bf164096256bf1640962048bf16
scrolls · 16 cases total
Aggregatessuite of 16 cases · mean latency
HardwareBest knownImplementationRuns
NVIDIA B200 · env 1261.2 µsRecursive17NVIDIA B200 · env 2250.9 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 · 17 results · last observed 2026-08-19Record history →
#
Implementation
Latency
vs #1
Trust
Observed
1
Recursive
261.2µs94% SOL
1.00×
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 →
2
Amir M. Mir | SF Tensor
280.7µs93% SOL
1.08×
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
Geometric
315.8µs92% SOL
1.21×
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 →
4
Databricks
316.9µs92% SOL
1.21×
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
Amir M. Mir | SF Tensor
333.9µs91% SOL
1.28×
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
342.9µs91% SOL
1.31×
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 →
7
TH3CHARLie | TileGym
347.7µs91% SOL
1.33×
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 →
8
ac4k
354.9µs91% 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 →
9
Geometric
366.4µs90% SOL
1.40×
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 →
10
Geometric
375.8µs90% SOL
1.44×
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 →
11
doubleAI
387.9µs89% SOL
1.49×
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
Geometric
475.9µs86% SOL
1.82×
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 →
13
Geometric
503.6µs85% SOL
1.93×
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 →
14
Geometric
519.2µs85% SOL
1.99×
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
TH3CHARLie | TileGym
606.3µs82% SOL
2.32×
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 →
16
Geometric
706.6µs80% SOL
2.71×
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 →
17
Cursor
1.10ms69% SOL
4.21×
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 →

Implementations

Implementation
Runtime
Best latency
Evidence
Availability
Recursive
unknown
250.9µs
1.00×
Reported
license unknown · no source
Databricks
unknown
294.7µs
1.17×
Reported
license unknown · no source
Geometric
unknown
315.8µs
1.26×
Reported
license unknown · no source
ac4k
unknown
342.3µs
1.36×
Reported
license unknown · no source
Hyra
unknown
342.9µs
1.37×
Reported
license unknown · no source
TH3CHARLie | TileGym
unknown
347.7µs
1.39×
Reported
license unknown · no source
Argus
unknown
353.4µs
1.41×
Reported
license unknown · no source
Cursor
unknown
889.6µs
3.55×
Reported
license unknown · no source
Show all 29 implementations ›
Zerg AI
unknown
5.26ms
20.97×
Reported
license unknown · no source
Igor Bragaia
unknown
5.46ms
21.77×
Reported
license unknown · no source
Databricks
unknown
316.9µs
1.26×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
317.9µs
1.27×
Reported
license unknown · no source
doubleAI
unknown
323.2µs
1.29×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
333.9µs
1.33×
Reported
license unknown · no source
ac4k
unknown
354.9µs
1.41×
Reported
license unknown · no source
Geometric
unknown
366.4µs
1.46×
Reported
license unknown · no source
Geometric
unknown
375.8µs
1.50×
Reported
license unknown · no source
doubleAI
unknown
387.9µs
1.55×
Reported
license unknown · no source
Geometric
unknown
475.9µs
1.90×
Reported
license unknown · no source
Geometric
unknown
503.6µs
2.01×
Reported
license unknown · no source
Geometric
unknown
519.2µs
2.07×
Reported
license unknown · no source
TH3CHARLie | TileGym
unknown
606.3µs
2.42×
Reported
license unknown · no source
Geometric
unknown
706.6µs
2.82×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
787.1µs
3.14×
Reported
license unknown · no source
Cursor
unknown
1.10ms
4.38×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
1.58ms
6.30×
Reported
license unknown · no source
Recursive
unknown
261.2µs
1.04×
Reported
license unknown · no source
doubleAI
unknown
279.9µs
1.12×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
280.7µs
1.12×
Reported
license unknown · no source

Semantics

Inputs and outputs
grad_outputbf16 [batch_seq_len, hidden_dim]
token_indicesint64 [num_tokens]
selected_tokensbf16 [num_tokens, hidden_dim]
w1_outputbf16 [num_tokens, ffn_dim]
gate_outputbf16 [num_tokens, ffn_dim]
up_outputbf16 [num_tokens, ffn_dim]
gated_outputbf16 [num_tokens, ffn_dim]
expert_outputbf16 [num_tokens, hidden_dim]
selected_weightsbf16 [num_tokens]
w1_weightbf16 [ffn_dim, hidden_dim]
w2_weightbf16 [hidden_dim, ffn_dim]
w3_weightbf16 [ffn_dim, hidden_dim]
grad_hidden_statesbf16 [batch_seq_len, hidden_dim]
grad_routing_weightsbf16 [batch_seq_len]
grad_w1_weightbf16 [ffn_dim, hidden_dim]
grad_w2_weightbf16 [hidden_dim, ffn_dim]
grad_w3_weightbf16 [ffn_dim, hidden_dim]
Axes and behavior
ffn_dimconstant = 14336
hidden_dimconstant = 4096
num_tokensvariable
batch_seq_lenvariable
determinismunspecified
constraintsNo mutation or aliasing
Identity
alias009_expert_token_scatter_with_weighted_forward_backwardmodelminimax-m2sha2569ca80a608133…
Sources: NVIDIA SOL-ExecBench (2026-08-19)last observed 2026-08-19How records are decidedJSON