# Token Throughput on LLaMA 8B 3.1 (4K-128K Context)

Source: [Token Throughput on LLaMA 8B 3.1 (4K-128K Context)](<https://www.sota2.com/research/sota/token-throughput-on-llama-8b-3-1-4k-128k-context>)

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

Data updated: 2026-04-23T09:18:19.726Z

Evaluation of the token throughput performance (tokens/sec) for the LLaMA 8B 3.1 model across various context lengths, specifically measuring performance at 4K, 16K, 32K, 64K, and 128K 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 | Token Throughput |
| Dataset (including any stated split) | LLaMA 8B 3.1 |
| Metric | Token Throughput (4K Context) (tokens/sec) |
| Unit | SCALAR |
| Direction | Higher is better |
| Results loaded | 5 of 5 |
| 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

- [Token Throughput (4K Context) (tokens/sec)](<https://www.sota2.com/research/sota/token-throughput-on-llama-8b-3-1-4k-128k-context.md?metric=Token%20Throughput%20(4K%20Context)%20(tokens%2Fsec)>): Higher is better; unit: SCALAR.

- [Token Throughput (16K Context) (tokens/sec)](<https://www.sota2.com/research/sota/token-throughput-on-llama-8b-3-1-4k-128k-context.md?metric=Token%20Throughput%20(16K%20Context)%20(tokens%2Fsec)>): Higher is better; unit: SCALAR.

- [Token Throughput (32K Context) (tokens/sec)](<https://www.sota2.com/research/sota/token-throughput-on-llama-8b-3-1-4k-128k-context.md?metric=Token%20Throughput%20(32K%20Context)%20(tokens%2Fsec)>): Higher is better; unit: SCALAR.

- [Token Throughput (64K Context) (tokens/sec)](<https://www.sota2.com/research/sota/token-throughput-on-llama-8b-3-1-4k-128k-context.md?metric=Token%20Throughput%20(64K%20Context)%20(tokens%2Fsec)>): Higher is better; unit: SCALAR.

- [Token Throughput (128K Context) (tokens/sec)](<https://www.sota2.com/research/sota/token-throughput-on-llama-8b-3-1-4k-128k-context.md?metric=Token%20Throughput%20(128K%20Context)%20(tokens%2Fsec)>): Higher is better; unit: SCALAR.

## Recorded results

| Method | Token Throughput (4K Context) (tokens/sec) | Result date | Report paper |
| --- | --- | --- | --- |
| ShadowKV | 150.8 | Not recorded | 69e9a76155a3c32af89f6ea2 |
| TTKV | 148.7 | Not recorded | 69e9a76155a3c32af89f6ea2 |
| KVQuant (3-bit) | 145.1 | Not recorded | 69e9a76155a3c32af89f6ea2 |
| KIVI (2-bit) | 142.3 | Not recorded | 69e9a76155a3c32af89f6ea2 |
| FP16 | 135.5 | Not recorded | 69e9a76155a3c32af89f6ea2 |

### ShadowKV

Token Throughput (4K Context) (tokens/sec): 150.8. Result ID: 69e9e3cc47f34ce924d14fec.

[Reporting paper](<https://www.sota2.com/research/paper/69e9a76155a3c32af89f6ea2.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78461e209131cea778ce.md>)

Recorded evaluation configuration: {"Batch size":"8"}. This record does not assert independent reproduction.

### TTKV

Token Throughput (4K Context) (tokens/sec): 148.7. Result ID: 69e9e3cc47f34ce924d14fed.

[Reporting paper](<https://www.sota2.com/research/paper/69e9a76155a3c32af89f6ea2.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69e9a76155a3c32af89f6ea2.md>)

Recorded evaluation configuration: {"Batch size":"8"}. This record does not assert independent reproduction.

### KVQuant (3-bit)

Token Throughput (4K Context) (tokens/sec): 145.1. Result ID: 69e9e3cc47f34ce924d14feb.

[Reporting paper](<https://www.sota2.com/research/paper/69e9a76155a3c32af89f6ea2.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78eb1e209131cea7b365.md>)

[Implementation](<https://github.com/SqueezeAILab/KVQuant/>)

Recorded evaluation configuration: {"Batch size":"8"}. This record does not assert independent reproduction.

### KIVI (2-bit)

Token Throughput (4K Context) (tokens/sec): 142.3. Result ID: 69e9e3cc47f34ce924d14fea.

[Reporting paper](<https://www.sota2.com/research/paper/69e9a76155a3c32af89f6ea2.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa744b1e209131cea60dab.md>)

Recorded evaluation configuration: {"Batch size":"8"}. This record does not assert independent reproduction.

### FP16

Token Throughput (4K Context) (tokens/sec): 135.5. Result ID: 69e9e3cc47f34ce924d14fe9.

[Reporting paper](<https://www.sota2.com/research/paper/69e9a76155a3c32af89f6ea2.md>)

Recorded evaluation configuration: {"Batch size":"8"}. This record does not assert independent reproduction.

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