# KV-Cache Footprint Analysis on Sequence length T=2048 BF16 (inference)

Source: [KV-Cache Footprint Analysis on Sequence length T=2048 BF16 (inference)](<https://www.sota2.com/research/sota/kv-cache-footprint-analysis-on-sequence-length-t-2048-bf16-inference>)

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

Data updated: 2026-07-03T06:26:48.444Z

Evaluation of the Key-Value (KV) cache memory footprint required for inference at a sequence length of 2048 tokens, measured using BF16 precision.

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 Footprint Analysis |
| Dataset (including any stated split) | Sequence length T=2048 BF16 (inference) |
| Metric | KV Cache Size (MB) |
| Unit | SIZE |
| Direction | Lower is better |
| Results loaded | 4 of 4 |
| 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

- [KV Cache Size (MB)](<https://www.sota2.com/research/sota/kv-cache-footprint-analysis-on-sequence-length-t-2048-bf16-inference.md?metric=KV%20Cache%20Size%20(MB)>): Lower is better; unit: SIZE.

## Recorded results

| Method | KV Cache Size (MB) | Result date | Report paper |
| --- | --- | --- | --- |
| Wiola-360m | 67 | Not recorded | 6a4741e2143ebdcacacc59c6 |
| OPT-350M | 168 | Not recorded | 6a4741e2143ebdcacacc59c6 |
| Pythia-410M | 192 | Not recorded | 6a4741e2143ebdcacacc59c6 |
| GPT-2 XL | 421 | Not recorded | 6a4741e2143ebdcacacc59c6 |

### Wiola-360m

KV Cache Size (MB): 67. Result ID: 6a47560a60b88bb33991c7ff.

[Reporting paper](<https://www.sota2.com/research/paper/6a4741e2143ebdcacacc59c6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a4741e2143ebdcacacc59c6.md>)

Recorded evaluation configuration: {"Params":"361M"}. This record does not assert independent reproduction.

### OPT-350M

KV Cache Size (MB): 168. Result ID: 6a47560a60b88bb33991c7fd.

[Reporting paper](<https://www.sota2.com/research/paper/6a4741e2143ebdcacacc59c6.md>)

Recorded evaluation configuration: {"Params":"350M"}. This record does not assert independent reproduction.

### Pythia-410M

KV Cache Size (MB): 192. Result ID: 6a47560a60b88bb33991c7fe.

[Reporting paper](<https://www.sota2.com/research/paper/6a4741e2143ebdcacacc59c6.md>)

Recorded evaluation configuration: {"Params":"410M"}. This record does not assert independent reproduction.

### GPT-2 XL

KV Cache Size (MB): 421. Result ID: 6a47560a60b88bb33991c7fc.

[Reporting paper](<https://www.sota2.com/research/paper/6a4741e2143ebdcacacc59c6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa680c1e209131cea19d53.md>)

[Implementation](<https://github.com/openai/gpt-3/blob/master/overlap_frequency.md>)

Recorded evaluation configuration: {"Params":"1.5B"}. This record does not assert independent reproduction.

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