# Sigmoid Attention on Llama2-7b (q=32, k=32) (1k)

Source: [Sigmoid Attention on Llama2-7b (q=32, k=32) (1k)](<https://www.sota2.com/research/sota/sigmoid-attention-on-llama2-7b-q-32-k-32-1k>)

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

Data updated: 2026-05-07T09:37:12.001Z

Measurement of the computational throughput (TFLOPS) for the Sigmoid Attention operation on the Llama 2 7B model, configured with q=32 and k=32 heads, evaluated using a context length of 1k tokens.

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 | Sigmoid Attention |
| Dataset (including any stated split) | Llama2-7b (q=32, k=32) (1k) |
| Metric | Throughput (TFLOPS) |
| Unit | SCALAR |
| Direction | Higher is better |
| Results loaded | 3 of 3 |
| 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

- [Throughput (TFLOPS)](<https://www.sota2.com/research/sota/sigmoid-attention-on-llama2-7b-q-32-k-32-1k.md?metric=Throughput%20(TFLOPS)>): Higher is better; unit: SCALAR.

## Recorded results

| Method | Throughput (TFLOPS) | Result date | Report paper |
| --- | --- | --- | --- |
| CuBridge | 134.56 | Not recorded | 69fc1c955497b3b670c88575 |
| Qimeng Attn | 45.05 | Not recorded | 69fc1c955497b3b670c88575 |
| Torch | 30.48 | Not recorded | 69fc1c955497b3b670c88575 |

### CuBridge

Throughput (TFLOPS): 134.56. Result ID: 69fc5d2f47f34ce924df7315.

[Reporting paper](<https://www.sota2.com/research/paper/69fc1c955497b3b670c88575.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69fc1c955497b3b670c88575.md>)

Recorded evaluation configuration: {"Implementation":"CuBridge"}. This record does not assert independent reproduction.

### Qimeng Attn

Throughput (TFLOPS): 45.05. Result ID: 69fc5d2f47f34ce924df7314.

[Reporting paper](<https://www.sota2.com/research/paper/69fc1c955497b3b670c88575.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa79671e209131cea7d6cb.md>)

Recorded evaluation configuration: {"Implementation":"Qimeng Attention"}. This record does not assert independent reproduction.

### Torch

Throughput (TFLOPS): 30.48. Result ID: 69fc5d2f47f34ce924df7313.

[Reporting paper](<https://www.sota2.com/research/paper/69fc1c955497b3b670c88575.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66cc1e209131cea1266b.md>)

Recorded evaluation configuration: {"Implementation":"Standard PyTorch"}. This record does not assert independent reproduction.

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