# Jacobian computation on Transformer

Source: [Jacobian computation on Transformer](<https://www.sota2.com/research/sota/jacobian-computation-on-transformer>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-02-26T05:52:34.590Z

Evaluation of algorithms performing Jacobian computation on the Transformer dataset, typically measured by median runtime.

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Jacobian computation |
| Dataset (including any stated split) | Transformer |
| Metric | Median Runtime (ms) |
| Unit | TIME |
| Direction | Lower is better |
| Results loaded | 8 of 8 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [Median Runtime (ms)](<https://www.sota2.com/research/sota/jacobian-computation-on-transformer.md?metric=Median%20Runtime%20(ms)>): Lower is better; unit: TIME.

## Recorded results

| Method | Median Runtime (ms) | Result date | Report paper |
| --- | --- | --- | --- |
| AlphaGrad | 0.28 | Not recorded | 68fa76b91e209131cea6ebac |
| Reverse | 0.29 | Not recorded | 68fa76b91e209131cea6ebac |
| Markowitz | 16.17 | Not recorded | 68fa76b91e209131cea6ebac |
| Forward | 21.38 | Not recorded | 68fa76b91e209131cea6ebac |
| Reverse | 39.07 | Not recorded | 68fa76b91e209131cea6ebac |
| AlphaGrad | 40.38 | Not recorded | 68fa76b91e209131cea6ebac |
| Markowitz | 498.94 | Not recorded | 68fa76b91e209131cea6ebac |
| Forward | 990.09 | Not recorded | 68fa76b91e209131cea6ebac |

### AlphaGrad

Median Runtime (ms): 0.28. Result ID: 69945f9d3a6c3148491cd5b4.

[Reporting paper](<https://www.sota2.com/research/paper/68fa76b91e209131cea6ebac.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76b91e209131cea6ebac.md>)

[Implementation](<https://github.com/jamielohoff/alphagrad>)

Recorded evaluation configuration: {"AD Mode":"AlphaGrad","Device":"NVIDIA RTX 4090 (GPU)","Batch Size":1,"Compilation":"JIT"}. This record does not assert independent reproduction.

### Reverse

Median Runtime (ms): 0.29. Result ID: 69945f9d3a6c3148491cd5b2.

[Reporting paper](<https://www.sota2.com/research/paper/68fa76b91e209131cea6ebac.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa63f51e209131cea01d42.md>)

Recorded evaluation configuration: {"AD Mode":"Reverse","Device":"NVIDIA RTX 4090 (GPU)","Batch Size":1,"Compilation":"JIT"}. This record does not assert independent reproduction.

### Markowitz

Median Runtime (ms): 16.17. Result ID: 69945f9d3a6c3148491cd5b3.

[Reporting paper](<https://www.sota2.com/research/paper/68fa76b91e209131cea6ebac.md>)

Recorded evaluation configuration: {"AD Mode":"Markowitz","Device":"NVIDIA RTX 4090 (GPU)","Batch Size":1,"Compilation":"JIT"}. This record does not assert independent reproduction.

### Forward

Median Runtime (ms): 21.38. Result ID: 69945f9d3a6c3148491cd5b1.

[Reporting paper](<https://www.sota2.com/research/paper/68fa76b91e209131cea6ebac.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa63f51e209131cea01d42.md>)

Recorded evaluation configuration: {"AD Mode":"Forward","Device":"NVIDIA RTX 4090 (GPU)","Batch Size":1,"Compilation":"JIT"}. This record does not assert independent reproduction.

### Reverse

Median Runtime (ms): 39.07. Result ID: 69945f9d3a6c3148491cd5ae.

[Reporting paper](<https://www.sota2.com/research/paper/68fa76b91e209131cea6ebac.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa63f51e209131cea01d42.md>)

Recorded evaluation configuration: {"AD Mode":"Reverse","Device":"AMD EPYC 9684X (CPU)","Batch Size":1,"Compilation":"XLA"}. This record does not assert independent reproduction.

### AlphaGrad

Median Runtime (ms): 40.38. Result ID: 69945f9d3a6c3148491cd5b0.

[Reporting paper](<https://www.sota2.com/research/paper/68fa76b91e209131cea6ebac.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76b91e209131cea6ebac.md>)

[Implementation](<https://github.com/jamielohoff/alphagrad>)

Recorded evaluation configuration: {"AD Mode":"AlphaGrad","Device":"AMD EPYC 9684X (CPU)","Batch Size":1,"Compilation":"XLA"}. This record does not assert independent reproduction.

### Markowitz

Median Runtime (ms): 498.94. Result ID: 69945f9d3a6c3148491cd5af.

[Reporting paper](<https://www.sota2.com/research/paper/68fa76b91e209131cea6ebac.md>)

Recorded evaluation configuration: {"AD Mode":"Markowitz","Device":"AMD EPYC 9684X (CPU)","Batch Size":1,"Compilation":"XLA"}. This record does not assert independent reproduction.

### Forward

Median Runtime (ms): 990.09. Result ID: 69945f9d3a6c3148491cd5ad.

[Reporting paper](<https://www.sota2.com/research/paper/68fa76b91e209131cea6ebac.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa63f51e209131cea01d42.md>)

Recorded evaluation configuration: {"AD Mode":"Forward","Device":"AMD EPYC 9684X (CPU)","Batch Size":1,"Compilation":"XLA"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.