# Inference Efficiency on DeepScaleR-40k (1,024 mathematical problems)

Source: [Inference Efficiency on DeepScaleR-40k (1,024 mathematical problems)](<https://www.sota2.com/research/sota/inference-efficiency-on-deepscaler-40k-1-024-mathematical-problems>)

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

Data updated: 2026-02-26T06:06:26.136Z

Evaluation of model inference efficiency, specifically measuring throughput (tokens/s), on the DeepScaleR-40k dataset containing 1,024 mathematical problems.

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 | Inference Efficiency |
| Dataset (including any stated split) | DeepScaleR-40k (1,024 mathematical problems) |
| Metric | Throughput (tokens/s) |
| Unit | TIME |
| Direction | Higher is better |
| Results loaded | 26 of 26 |
| 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 (tokens/s)](<https://www.sota2.com/research/sota/inference-efficiency-on-deepscaler-40k-1-024-mathematical-problems.md?metric=Throughput%20(tokens%2Fs)>): Higher is better; unit: TIME.

## Recorded results

| Method | Throughput (tokens/s) | Result date | Report paper |
| --- | --- | --- | --- |
| G-KV | 760.74 | Not recorded | 692e72a5d6b0dd33429323a0 |
| G-KV | 612.6 | Not recorded | 692e72a5d6b0dd33429323a0 |
| G-KV | 517.3 | Not recorded | 692e72a5d6b0dd33429323a0 |
| G-KV | 475.59 | Not recorded | 692e72a5d6b0dd33429323a0 |
| G-KV | 448.35 | Not recorded | 692e72a5d6b0dd33429323a0 |
| G-KV | 367.96 | Not recorded | 692e72a5d6b0dd33429323a0 |
| G-KV | 261.32 | Not recorded | 692e72a5d6b0dd33429323a0 |
| G-KV | 258.18 | Not recorded | 692e72a5d6b0dd33429323a0 |
| G-KV | 248.22 | Not recorded | 692e72a5d6b0dd33429323a0 |
| R-KV | 238.43 | Not recorded | 692e72a5d6b0dd33429323a0 |
| G-KV | 221.23 | Not recorded | 692e72a5d6b0dd33429323a0 |
| G-KV | 212.93 | Not recorded | 692e72a5d6b0dd33429323a0 |
| R-KV | 203.66 | Not recorded | 692e72a5d6b0dd33429323a0 |
| G-KV | 193.56 | Not recorded | 692e72a5d6b0dd33429323a0 |
| R-KV | 172.44 | Not recorded | 692e72a5d6b0dd33429323a0 |
| G-KV | 170.64 | Not recorded | 692e72a5d6b0dd33429323a0 |
| G-KV | 158.43 | Not recorded | 692e72a5d6b0dd33429323a0 |
| G-KV | 154.52 | Not recorded | 692e72a5d6b0dd33429323a0 |
| G-KV | 118.82 | Not recorded | 692e72a5d6b0dd33429323a0 |
| G-KV | 118.1 | Not recorded | 692e72a5d6b0dd33429323a0 |
| R-KV | 111.89 | Not recorded | 692e72a5d6b0dd33429323a0 |
| R-KV | 98.01 | Not recorded | 692e72a5d6b0dd33429323a0 |
| G-KV | 93.91 | Not recorded | 692e72a5d6b0dd33429323a0 |
| R-KV | 82.33 | Not recorded | 692e72a5d6b0dd33429323a0 |
| Full-KV | 62.41 | Not recorded | 692e72a5d6b0dd33429323a0 |
| Full-KV | 31.08 | Not recorded | 692e72a5d6b0dd33429323a0 |

### G-KV

Throughput (tokens/s): 760.74. Result ID: 6996d4785698665ec420cdda.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Qwen-Distill-7B","Batch Size":"128","KV Cache Budget":"512"}. This record does not assert independent reproduction.

### G-KV

Throughput (tokens/s): 612.6. Result ID: 6996d4785698665ec420cde7.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Llama-Distill-8B","Batch Size":"64","KV Cache Budget":"512"}. This record does not assert independent reproduction.

### G-KV

Throughput (tokens/s): 517.3. Result ID: 6996d4785698665ec420cde6.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Llama-Distill-8B","Batch Size":"32","KV Cache Budget":"512"}. This record does not assert independent reproduction.

### G-KV

Throughput (tokens/s): 475.59. Result ID: 6996d4785698665ec420cdd9.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Qwen-Distill-7B","Batch Size":"64","KV Cache Budget":"512"}. This record does not assert independent reproduction.

### G-KV

Throughput (tokens/s): 448.35. Result ID: 6996d4785698665ec420cddd.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Qwen-Distill-7B","Batch Size":"128","KV Cache Budget":"1024"}. This record does not assert independent reproduction.

### G-KV

Throughput (tokens/s): 367.96. Result ID: 6996d4785698665ec420cddc.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Qwen-Distill-7B","Batch Size":"64","KV Cache Budget":"1024"}. This record does not assert independent reproduction.

### G-KV

Throughput (tokens/s): 261.32. Result ID: 6996d4785698665ec420cdd8.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Qwen-Distill-7B","Batch Size":"32","KV Cache Budget":"512"}. This record does not assert independent reproduction.

### G-KV

Throughput (tokens/s): 258.18. Result ID: 6996d4785698665ec420cdea.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Llama-Distill-8B","Batch Size":"64","KV Cache Budget":"1024"}. This record does not assert independent reproduction.

### G-KV

Throughput (tokens/s): 248.22. Result ID: 6996d4785698665ec420cde0.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Qwen-Distill-7B","Batch Size":"128","KV Cache Budget":"2048"}. This record does not assert independent reproduction.

### R-KV

Throughput (tokens/s): 238.43. Result ID: 6996d4785698665ec420cdd7.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Qwen-Distill-7B","Batch Size":"128","KV Cache Budget":"2048"}. This record does not assert independent reproduction.

### G-KV

Throughput (tokens/s): 221.23. Result ID: 6996d4785698665ec420cddf.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Qwen-Distill-7B","Batch Size":"64","KV Cache Budget":"2048"}. This record does not assert independent reproduction.

### G-KV

Throughput (tokens/s): 212.93. Result ID: 6996d4785698665ec420cddb.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Qwen-Distill-7B","Batch Size":"32","KV Cache Budget":"1024"}. This record does not assert independent reproduction.

### R-KV

Throughput (tokens/s): 203.66. Result ID: 6996d4785698665ec420cdd6.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Qwen-Distill-7B","Batch Size":"64","KV Cache Budget":"2048"}. This record does not assert independent reproduction.

### G-KV

Throughput (tokens/s): 193.56. Result ID: 6996d4785698665ec420cde9.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Llama-Distill-8B","Batch Size":"32","KV Cache Budget":"1024"}. This record does not assert independent reproduction.

### R-KV

Throughput (tokens/s): 172.44. Result ID: 6996d4785698665ec420cdd5.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Qwen-Distill-7B","Batch Size":"32","KV Cache Budget":"2048"}. This record does not assert independent reproduction.

### G-KV

Throughput (tokens/s): 170.64. Result ID: 6996d4785698665ec420cdde.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Qwen-Distill-7B","Batch Size":"32","KV Cache Budget":"2048"}. This record does not assert independent reproduction.

### G-KV

Throughput (tokens/s): 158.43. Result ID: 6996d4785698665ec420cde5.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Llama-Distill-8B","Batch Size":"16","KV Cache Budget":"512"}. This record does not assert independent reproduction.

### G-KV

Throughput (tokens/s): 154.52. Result ID: 6996d4785698665ec420cded.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Llama-Distill-8B","Batch Size":"64","KV Cache Budget":"2048"}. This record does not assert independent reproduction.

### G-KV

Throughput (tokens/s): 118.82. Result ID: 6996d4785698665ec420cdec.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Llama-Distill-8B","Batch Size":"32","KV Cache Budget":"2048"}. This record does not assert independent reproduction.

### G-KV

Throughput (tokens/s): 118.1. Result ID: 6996d4785698665ec420cde8.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Llama-Distill-8B","Batch Size":"16","KV Cache Budget":"1024"}. This record does not assert independent reproduction.

### R-KV

Throughput (tokens/s): 111.89. Result ID: 6996d4785698665ec420cde4.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Llama-Distill-8B","Batch Size":"64","KV Cache Budget":"2048"}. This record does not assert independent reproduction.

### R-KV

Throughput (tokens/s): 98.01. Result ID: 6996d4785698665ec420cde3.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Llama-Distill-8B","Batch Size":"32","KV Cache Budget":"2048"}. This record does not assert independent reproduction.

### G-KV

Throughput (tokens/s): 93.91. Result ID: 6996d4785698665ec420cdeb.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Llama-Distill-8B","Batch Size":"16","KV Cache Budget":"2048"}. This record does not assert independent reproduction.

### R-KV

Throughput (tokens/s): 82.33. Result ID: 6996d4785698665ec420cde2.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Llama-Distill-8B","Batch Size":"16","KV Cache Budget":"2048"}. This record does not assert independent reproduction.

### Full-KV

Throughput (tokens/s): 62.41. Result ID: 6996d4785698665ec420cdd4.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Qwen-Distill-7B","Batch Size":"32"}. This record does not assert independent reproduction.

### Full-KV

Throughput (tokens/s): 31.08. Result ID: 6996d4785698665ec420cde1.

[Reporting paper](<https://www.sota2.com/research/paper/692e72a5d6b0dd33429323a0.md>)

Recorded evaluation configuration: {"Model":"DeepSeek-Llama-Distill-8B","Batch Size":"16"}. This record does not assert independent reproduction.

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