# KV Cache Compression on 5 sequences Qwen2.5-1.5B-Instruct

Source: [KV Cache Compression on 5 sequences Qwen2.5-1.5B-Instruct](<https://www.sota2.com/research/sota/kv-cache-compression-on-5-sequences-qwen2-5-1-5b-instruct>)

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

Data updated: 2026-03-31T10:31:06.708Z

Evaluation of model performance on KV Cache Compression using the '5 sequences Qwen2.5-1.5B-Instruct' context, measured by KL divergence.

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 | KV Cache Compression |
| Dataset (including any stated split) | 5 sequences Qwen2.5-1.5B-Instruct |
| Metric | KL Divergence |
| Unit | SCALAR |
| Direction | Lower is better |
| Results loaded | 15 of 15 |
| 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

- [KL Divergence](<https://www.sota2.com/research/sota/kv-cache-compression-on-5-sequences-qwen2-5-1-5b-instruct.md?metric=KL%20Divergence>): Lower is better; unit: SCALAR.

## Recorded results

| Method | KL Divergence | Result date | Report paper |
| --- | --- | --- | --- |
| KVSCULPT | 0.0358 | Not recorded | 69cb552e0636e5bf4afca028 |
| KVSCULPT | 0.0463 | Not recorded | 69cb552e0636e5bf4afca028 |
| KVSCULPT | 0.0575 | Not recorded | 69cb552e0636e5bf4afca028 |
| Joint Optimization | 0.115 | Not recorded | 69cb552e0636e5bf4afca028 |
| Select+Fit | 0.125 | Not recorded | 69cb552e0636e5bf4afca028 |
| Attention Score | 0.139 | Not recorded | 69cb552e0636e5bf4afca028 |
| Joint Optimization | 0.18 | Not recorded | 69cb552e0636e5bf4afca028 |
| Select+Fit | 0.186 | Not recorded | 69cb552e0636e5bf4afca028 |
| Attention Score | 0.204 | Not recorded | 69cb552e0636e5bf4afca028 |
| Joint Optimization | 0.224 | Not recorded | 69cb552e0636e5bf4afca028 |
| Select+Fit | 0.233 | Not recorded | 69cb552e0636e5bf4afca028 |
| Attention Score | 0.254 | Not recorded | 69cb552e0636e5bf4afca028 |
| Random | 1.54 | Not recorded | 69cb552e0636e5bf4afca028 |
| Random | 1.8 | Not recorded | 69cb552e0636e5bf4afca028 |
| Random | 2.5 | Not recorded | 69cb552e0636e5bf4afca028 |

### KVSCULPT

KL Divergence: 0.0358. Result ID: 69cba11f47f34ce924b48714.

[Reporting paper](<https://www.sota2.com/research/paper/69cb552e0636e5bf4afca028.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69cb552e0636e5bf4afca028.md>)

Recorded evaluation configuration: {"Compression Ratio (r)":"0.7","Retain zone (m)":"256","Optimizer":"L-BFGS","Learning rate":"0.5"}. This record does not assert independent reproduction.

### KVSCULPT

KL Divergence: 0.0463. Result ID: 69cba11f47f34ce924b48713.

[Reporting paper](<https://www.sota2.com/research/paper/69cb552e0636e5bf4afca028.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69cb552e0636e5bf4afca028.md>)

Recorded evaluation configuration: {"Compression Ratio (r)":"0.5","Retain zone (m)":"256","Optimizer":"L-BFGS","Learning rate":"0.5"}. This record does not assert independent reproduction.

### KVSCULPT

KL Divergence: 0.0575. Result ID: 69cba11f47f34ce924b48712.

[Reporting paper](<https://www.sota2.com/research/paper/69cb552e0636e5bf4afca028.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69cb552e0636e5bf4afca028.md>)

Recorded evaluation configuration: {"Compression Ratio (r)":"0.3","Retain zone (m)":"256","Optimizer":"L-BFGS","Learning rate":"0.5"}. This record does not assert independent reproduction.

### Joint Optimization

KL Divergence: 0.115. Result ID: 69cba11f47f34ce924b48711.

[Reporting paper](<https://www.sota2.com/research/paper/69cb552e0636e5bf4afca028.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa647d1e209131cea04d19.md>)

Recorded evaluation configuration: {"Compression Ratio (r)":"0.7","Retain zone (m)":"256"}. This record does not assert independent reproduction.

### Select+Fit

KL Divergence: 0.125. Result ID: 69cba11f47f34ce924b4870e.

[Reporting paper](<https://www.sota2.com/research/paper/69cb552e0636e5bf4afca028.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75901e209131cea68a91.md>)

Recorded evaluation configuration: {"Compression Ratio (r)":"0.7","Retain zone (m)":"256"}. This record does not assert independent reproduction.

### Attention Score

KL Divergence: 0.139. Result ID: 69cba11f47f34ce924b4870b.

[Reporting paper](<https://www.sota2.com/research/paper/69cb552e0636e5bf4afca028.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa718c1e209131cea51ead.md>)

[Implementation](<https://github.com/FMInference/H2O>)

Recorded evaluation configuration: {"Compression Ratio (r)":"0.7","Retain zone (m)":"256"}. This record does not assert independent reproduction.

### Joint Optimization

KL Divergence: 0.18. Result ID: 69cba11f47f34ce924b48710.

[Reporting paper](<https://www.sota2.com/research/paper/69cb552e0636e5bf4afca028.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa647d1e209131cea04d19.md>)

Recorded evaluation configuration: {"Compression Ratio (r)":"0.5","Retain zone (m)":"256"}. This record does not assert independent reproduction.

### Select+Fit

KL Divergence: 0.186. Result ID: 69cba11f47f34ce924b4870d.

[Reporting paper](<https://www.sota2.com/research/paper/69cb552e0636e5bf4afca028.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75901e209131cea68a91.md>)

Recorded evaluation configuration: {"Compression Ratio (r)":"0.5","Retain zone (m)":"256"}. This record does not assert independent reproduction.

### Attention Score

KL Divergence: 0.204. Result ID: 69cba11f47f34ce924b4870a.

[Reporting paper](<https://www.sota2.com/research/paper/69cb552e0636e5bf4afca028.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa718c1e209131cea51ead.md>)

[Implementation](<https://github.com/FMInference/H2O>)

Recorded evaluation configuration: {"Compression Ratio (r)":"0.5","Retain zone (m)":"256"}. This record does not assert independent reproduction.

### Joint Optimization

KL Divergence: 0.224. Result ID: 69cba11f47f34ce924b4870f.

[Reporting paper](<https://www.sota2.com/research/paper/69cb552e0636e5bf4afca028.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa647d1e209131cea04d19.md>)

Recorded evaluation configuration: {"Compression Ratio (r)":"0.3","Retain zone (m)":"256"}. This record does not assert independent reproduction.

### Select+Fit

KL Divergence: 0.233. Result ID: 69cba11f47f34ce924b4870c.

[Reporting paper](<https://www.sota2.com/research/paper/69cb552e0636e5bf4afca028.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75901e209131cea68a91.md>)

Recorded evaluation configuration: {"Compression Ratio (r)":"0.3","Retain zone (m)":"256"}. This record does not assert independent reproduction.

### Attention Score

KL Divergence: 0.254. Result ID: 69cba11f47f34ce924b48709.

[Reporting paper](<https://www.sota2.com/research/paper/69cb552e0636e5bf4afca028.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa718c1e209131cea51ead.md>)

[Implementation](<https://github.com/FMInference/H2O>)

Recorded evaluation configuration: {"Compression Ratio (r)":"0.3","Retain zone (m)":"256"}. This record does not assert independent reproduction.

### Random

KL Divergence: 1.54. Result ID: 69cba11f47f34ce924b48708.

[Reporting paper](<https://www.sota2.com/research/paper/69cb552e0636e5bf4afca028.md>)

Recorded evaluation configuration: {"Compression Ratio (r)":"0.7","Retain zone (m)":"256"}. This record does not assert independent reproduction.

### Random

KL Divergence: 1.8. Result ID: 69cba11f47f34ce924b48707.

[Reporting paper](<https://www.sota2.com/research/paper/69cb552e0636e5bf4afca028.md>)

Recorded evaluation configuration: {"Compression Ratio (r)":"0.5","Retain zone (m)":"256"}. This record does not assert independent reproduction.

### Random

KL Divergence: 2.5. Result ID: 69cba11f47f34ce924b48706.

[Reporting paper](<https://www.sota2.com/research/paper/69cb552e0636e5bf4afca028.md>)

Recorded evaluation configuration: {"Compression Ratio (r)":"0.3","Retain zone (m)":"256"}. This record does not assert independent reproduction.

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