# Long-context language understanding on LongBench (Comprehensive Metric Suite)

Source: [Long-context language understanding on LongBench (Comprehensive Metric Suite)](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-comprehensive-metric-suite>)

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

Data updated: 2026-06-30T07:42:43.639Z

Comprehensive evaluation of long-context language understanding capabilities across a wide range of LongBench subtasks, including NrtvQA, Qasper, HotpotQA, TriviaQA, SAMSum, and others.

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 | Long-context language understanding |
| Dataset (including any stated split) | LongBench |
| Metric | NrtvQA Score |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 37 of 37 |
| 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

- [NrtvQA Score](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-comprehensive-metric-suite.md?metric=NrtvQA%20Score>): Higher is better; unit: PERCENT.

- [Qasper Score](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-comprehensive-metric-suite.md?metric=Qasper%20Score>): Higher is better; unit: PERCENT.

- [MF-en Score](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-comprehensive-metric-suite.md?metric=MF-en%20Score>): Higher is better; unit: PERCENT.

- [HotpotQA Score](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-comprehensive-metric-suite.md?metric=HotpotQA%20Score>): Higher is better; unit: PERCENT.

- [2WikiMQA Score](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-comprehensive-metric-suite.md?metric=2WikiMQA%20Score>): Higher is better; unit: PERCENT.

- [Musique Score](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-comprehensive-metric-suite.md?metric=Musique%20Score>): Higher is better; unit: PERCENT.

- [GovRep Score](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-comprehensive-metric-suite.md?metric=GovRep%20Score>): Higher is better; unit: PERCENT.

- [QMSum Score](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-comprehensive-metric-suite.md?metric=QMSum%20Score>): Higher is better; unit: PERCENT.

- [MultiNews Score](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-comprehensive-metric-suite.md?metric=MultiNews%20Score>): Higher is better; unit: PERCENT.

- [TREC Score](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-comprehensive-metric-suite.md?metric=TREC%20Score>): Higher is better; unit: PERCENT.

- [TriviaQA Score](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-comprehensive-metric-suite.md?metric=TriviaQA%20Score>): Higher is better; unit: PERCENT.

- [SAMSum Score](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-comprehensive-metric-suite.md?metric=SAMSum%20Score>): Higher is better; unit: PERCENT.

- [PCount](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-comprehensive-metric-suite.md?metric=PCount>): Higher is better; unit: SCALAR.

- [PRe Score](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-comprehensive-metric-suite.md?metric=PRe%20Score>): Higher is better; unit: PERCENT.

- [Lcc Score](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-comprehensive-metric-suite.md?metric=Lcc%20Score>): Higher is better; unit: PERCENT.

- [RB-P Score](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-comprehensive-metric-suite.md?metric=RB-P%20Score>): Higher is better; unit: PERCENT.

- [Average Score](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-comprehensive-metric-suite.md?metric=Average%20Score>): Higher is better; unit: PERCENT.

## Recorded results

| Method | NrtvQA Score | Result date | Report paper |
| --- | --- | --- | --- |
| K-VEC | 30.04 | Not recorded | 6a434d80ba73165dbb7439cb |
| Pyramid | 29.62 | Not recorded | 6a434d80ba73165dbb7439cb |
| SnapKV | 29.28 | Not recorded | 6a434d80ba73165dbb7439cb |
| Ada-SnapKV | 29.23 | Not recorded | 6a434d80ba73165dbb7439cb |
| Ada-Pyramid | 28.76 | Not recorded | 6a434d80ba73165dbb7439cb |
| FullKV | 27.96 | Not recorded | 6a02b3f7dbc836b87d824593 |
| SnapKV | 27.8 | Not recorded | 6a02b3f7dbc836b87d824593 |
| RDKV | 27.75 | Not recorded | 6a02b3f7dbc836b87d824593 |
| AdaKV | 27.64 | Not recorded | 6a02b3f7dbc836b87d824593 |
| RDKV | 27.35 | Not recorded | 6a02b3f7dbc836b87d824593 |
| RDKV | 26.79 | Not recorded | 6a02b3f7dbc836b87d824593 |
| ThinK | 26.73 | Not recorded | 6a02b3f7dbc836b87d824593 |
| AdaKV | 26.72 | Not recorded | 6a02b3f7dbc836b87d824593 |
| Snap+Zip | 26.6 | Not recorded | 6a02b3f7dbc836b87d824593 |
| SnapKV | 26.57 | Not recorded | 6a02b3f7dbc836b87d824593 |
| ThinK | 26.48 | Not recorded | 6a02b3f7dbc836b87d824593 |
| RDKV | 26.45 | Not recorded | 6a02b3f7dbc836b87d824593 |
| SnapKV | 25.92 | Not recorded | 6a02b3f7dbc836b87d824593 |
| Snap+Zip | 25.47 | Not recorded | 6a02b3f7dbc836b87d824593 |
| AdaKV | 25.46 | Not recorded | 6a02b3f7dbc836b87d824593 |
| ThinK | 25.3 | Not recorded | 6a02b3f7dbc836b87d824593 |
| SLM | 24.97 | Not recorded | 6a434d80ba73165dbb7439cb |
| RDKV | 24.94 | Not recorded | 6a02b3f7dbc836b87d824593 |
| K-VEC | 24.55 | Not recorded | 6a434d80ba73165dbb7439cb |
| Snap+Zip | 24.21 | Not recorded | 6a02b3f7dbc836b87d824593 |
| SnapKV | 23.88 | Not recorded | 6a434d80ba73165dbb7439cb |
| Snap+Zip | 23.84 | Not recorded | 6a02b3f7dbc836b87d824593 |
| AdaKV | 23.78 | Not recorded | 6a02b3f7dbc836b87d824593 |
| K-VEC | 23.45 | Not recorded | 6a434d80ba73165dbb7439cb |
| ThinK | 22.88 | Not recorded | 6a02b3f7dbc836b87d824593 |
| SnapKV | 21.46 | Not recorded | 6a02b3f7dbc836b87d824593 |
| GemFilter | 20.71 | Not recorded | 6a434d80ba73165dbb7439cb |
| SnapKV | 20.42 | Not recorded | 6a434d80ba73165dbb7439cb |
| ThinK | 18.65 | Not recorded | 6a02b3f7dbc836b87d824593 |
| Snap+Zip | 18.22 | Not recorded | 6a02b3f7dbc836b87d824593 |
| SnapKV | 16.95 | Not recorded | 6a02b3f7dbc836b87d824593 |
| AdaKV | 16.83 | Not recorded | 6a02b3f7dbc836b87d824593 |

### K-VEC

NrtvQA Score: 30.04. Result ID: 6a437359fbc373592c93c40b.

[Reporting paper](<https://www.sota2.com/research/paper/6a434d80ba73165dbb7439cb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a434d80ba73165dbb7439cb.md>)

Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct","Cache size":"1024"}. This record does not assert independent reproduction.

### Pyramid

NrtvQA Score: 29.62. Result ID: 6a437359fbc373592c93c406.

[Reporting paper](<https://www.sota2.com/research/paper/6a434d80ba73165dbb7439cb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78841e209131cea78cba.md>)

Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct","Cache size":"1024"}. This record does not assert independent reproduction.

### SnapKV

NrtvQA Score: 29.28. Result ID: 6a437359fbc373592c93c407.

[Reporting paper](<https://www.sota2.com/research/paper/6a434d80ba73165dbb7439cb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73b21e209131cea5df89.md>)

[Implementation](<https://github.com/FasterDecoding/SnapKV>)

Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct","Cache size":"1024"}. This record does not assert independent reproduction.

### Ada-SnapKV

NrtvQA Score: 29.23. Result ID: 6a437359fbc373592c93c409.

[Reporting paper](<https://www.sota2.com/research/paper/6a434d80ba73165dbb7439cb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c2e1e209131cea8ce46.md>)

Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct","Cache size":"1024"}. This record does not assert independent reproduction.

### Ada-Pyramid

NrtvQA Score: 28.76. Result ID: 6a437359fbc373592c93c408.

[Reporting paper](<https://www.sota2.com/research/paper/6a434d80ba73165dbb7439cb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c2e1e209131cea8ce46.md>)

Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct","Cache size":"1024"}. This record does not assert independent reproduction.

### FullKV

NrtvQA Score: 27.96. Result ID: 6a02dc2c47f34ce924e56aad.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"Full"}. This record does not assert independent reproduction.

### SnapKV

NrtvQA Score: 27.8. Result ID: 6a02dc2c47f34ce924e56ac2.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73b21e209131cea5df89.md>)

[Implementation](<https://github.com/FasterDecoding/SnapKV>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"1024L"}. This record does not assert independent reproduction.

### RDKV

NrtvQA Score: 27.75. Result ID: 6a02dc2c47f34ce924e56ac1.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"512L"}. This record does not assert independent reproduction.

### AdaKV

NrtvQA Score: 27.64. Result ID: 6a02dc2c47f34ce924e56ac3.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c2e1e209131cea8ce46.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"1024L"}. This record does not assert independent reproduction.

### RDKV

NrtvQA Score: 27.35. Result ID: 6a02dc2c47f34ce924e56ac6.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"1024L"}. This record does not assert independent reproduction.

### RDKV

NrtvQA Score: 26.79. Result ID: 6a02dc2c47f34ce924e56ab7.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"128L"}. This record does not assert independent reproduction.

### ThinK

NrtvQA Score: 26.73. Result ID: 6a02dc2c47f34ce924e56ac4.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa772c1e209131cea7190c.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"1024L"}. This record does not assert independent reproduction.

### AdaKV

NrtvQA Score: 26.72. Result ID: 6a02dc2c47f34ce924e56abe.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c2e1e209131cea8ce46.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"512L"}. This record does not assert independent reproduction.

### Snap+Zip

NrtvQA Score: 26.6. Result ID: 6a02dc2c47f34ce924e56ac5.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"1024L"}. This record does not assert independent reproduction.

### SnapKV

NrtvQA Score: 26.57. Result ID: 6a02dc2c47f34ce924e56abd.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73b21e209131cea5df89.md>)

[Implementation](<https://github.com/FasterDecoding/SnapKV>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"512L"}. This record does not assert independent reproduction.

### ThinK

NrtvQA Score: 26.48. Result ID: 6a02dc2c47f34ce924e56abf.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa772c1e209131cea7190c.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"512L"}. This record does not assert independent reproduction.

### RDKV

NrtvQA Score: 26.45. Result ID: 6a02dc2c47f34ce924e56abc.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"256L"}. This record does not assert independent reproduction.

### SnapKV

NrtvQA Score: 25.92. Result ID: 6a02dc2c47f34ce924e56ab8.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73b21e209131cea5df89.md>)

[Implementation](<https://github.com/FasterDecoding/SnapKV>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"256L"}. This record does not assert independent reproduction.

### Snap+Zip

NrtvQA Score: 25.47. Result ID: 6a02dc2c47f34ce924e56ac0.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"512L"}. This record does not assert independent reproduction.

### AdaKV

NrtvQA Score: 25.46. Result ID: 6a02dc2c47f34ce924e56ab9.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c2e1e209131cea8ce46.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"256L"}. This record does not assert independent reproduction.

### ThinK

NrtvQA Score: 25.3. Result ID: 6a02dc2c47f34ce924e56aba.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa772c1e209131cea7190c.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"256L"}. This record does not assert independent reproduction.

### SLM

NrtvQA Score: 24.97. Result ID: 6a437359fbc373592c93c405.

[Reporting paper](<https://www.sota2.com/research/paper/6a434d80ba73165dbb7439cb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72d71e209131cea58d79.md>)

Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct","Cache size":"1024"}. This record does not assert independent reproduction.

### RDKV

NrtvQA Score: 24.94. Result ID: 6a02dc2c47f34ce924e56ab2.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"64L"}. This record does not assert independent reproduction.

### K-VEC

NrtvQA Score: 24.55. Result ID: 6a437359fbc373592c93c412.

[Reporting paper](<https://www.sota2.com/research/paper/6a434d80ba73165dbb7439cb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a434d80ba73165dbb7439cb.md>)

Recorded evaluation configuration: {"Base Model":"Qwen2.5-7B-Instruct","Budget (B)":"1024"}. This record does not assert independent reproduction.

### Snap+Zip

NrtvQA Score: 24.21. Result ID: 6a02dc2c47f34ce924e56abb.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"256L"}. This record does not assert independent reproduction.

### SnapKV

NrtvQA Score: 23.88. Result ID: 6a437359fbc373592c93c411.

[Reporting paper](<https://www.sota2.com/research/paper/6a434d80ba73165dbb7439cb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73b21e209131cea5df89.md>)

[Implementation](<https://github.com/FasterDecoding/SnapKV>)

Recorded evaluation configuration: {"Base Model":"Qwen2.5-7B-Instruct","Budget (B)":"1024"}. This record does not assert independent reproduction.

### Snap+Zip

NrtvQA Score: 23.84. Result ID: 6a02dc2c47f34ce924e56ab6.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"128L"}. This record does not assert independent reproduction.

### AdaKV

NrtvQA Score: 23.78. Result ID: 6a02dc2c47f34ce924e56ab4.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c2e1e209131cea8ce46.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"128L"}. This record does not assert independent reproduction.

### K-VEC

NrtvQA Score: 23.45. Result ID: 6a437359fbc373592c93c410.

[Reporting paper](<https://www.sota2.com/research/paper/6a434d80ba73165dbb7439cb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a434d80ba73165dbb7439cb.md>)

Recorded evaluation configuration: {"Base Model":"Qwen2.5-7B-Instruct","Budget (B)":"128"}. This record does not assert independent reproduction.

### ThinK

NrtvQA Score: 22.88. Result ID: 6a02dc2c47f34ce924e56ab5.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa772c1e209131cea7190c.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"128L"}. This record does not assert independent reproduction.

### SnapKV

NrtvQA Score: 21.46. Result ID: 6a02dc2c47f34ce924e56ab3.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73b21e209131cea5df89.md>)

[Implementation](<https://github.com/FasterDecoding/SnapKV>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"128L"}. This record does not assert independent reproduction.

### GemFilter

NrtvQA Score: 20.71. Result ID: 6a437359fbc373592c93c40a.

[Reporting paper](<https://www.sota2.com/research/paper/6a434d80ba73165dbb7439cb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74fe1e209131cea651d5.md>)

Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct","Cache size":"1024"}. This record does not assert independent reproduction.

### SnapKV

NrtvQA Score: 20.42. Result ID: 6a437359fbc373592c93c40f.

[Reporting paper](<https://www.sota2.com/research/paper/6a434d80ba73165dbb7439cb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73b21e209131cea5df89.md>)

[Implementation](<https://github.com/FasterDecoding/SnapKV>)

Recorded evaluation configuration: {"Base Model":"Qwen2.5-7B-Instruct","Budget (B)":"128"}. This record does not assert independent reproduction.

### ThinK

NrtvQA Score: 18.65. Result ID: 6a02dc2c47f34ce924e56ab0.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa772c1e209131cea7190c.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"64L"}. This record does not assert independent reproduction.

### Snap+Zip

NrtvQA Score: 18.22. Result ID: 6a02dc2c47f34ce924e56ab1.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"64L"}. This record does not assert independent reproduction.

### SnapKV

NrtvQA Score: 16.95. Result ID: 6a02dc2c47f34ce924e56aae.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73b21e209131cea5df89.md>)

[Implementation](<https://github.com/FasterDecoding/SnapKV>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"64L"}. This record does not assert independent reproduction.

### AdaKV

NrtvQA Score: 16.83. Result ID: 6a02dc2c47f34ce924e56aaf.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b3f7dbc836b87d824593.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c2e1e209131cea8ce46.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Cache Budget (Btotal)":"64L"}. This record does not assert independent reproduction.

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