# Multi-step LLM Interaction on DeepSeek-R1-Distill-Llama-8B Inference Performance (test)

Source: [Multi-step LLM Interaction on DeepSeek-R1-Distill-Llama-8B Inference Performance (test)](<https://www.sota2.com/research/sota/multi-step-llm-interaction-on-deepseek-r1-distill-llama-8b-inference-performance-test>)

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

Data updated: 2026-02-26T05:46:27.465Z

Evaluation of the DeepSeek-R1-Distill-Llama-8B model's capability in performing multi-step interactions, measured using metrics such as High-Precision Ratio, Success Rate, and GHC.

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 | Multi-step LLM Interaction |
| Dataset (including any stated split) | DeepSeek-R1-Distill-Llama-8B Inference Performance (test) |
| Metric | High-Precision Ratio |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 7 of 7 |
| 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

- [High-Precision Ratio](<https://www.sota2.com/research/sota/multi-step-llm-interaction-on-deepseek-r1-distill-llama-8b-inference-performance-test.md?metric=High-Precision%20Ratio>): Higher is better; unit: PERCENT.

- [Success Rate](<https://www.sota2.com/research/sota/multi-step-llm-interaction-on-deepseek-r1-distill-llama-8b-inference-performance-test.md?metric=Success%20Rate>): Higher is better; unit: PERCENT.

- [GHC](<https://www.sota2.com/research/sota/multi-step-llm-interaction-on-deepseek-r1-distill-llama-8b-inference-performance-test.md?metric=GHC>): Higher is better; unit: SCALAR.

## Recorded results

| Method | High-Precision Ratio | Result date | Report paper |
| --- | --- | --- | --- |
| BF16 | 100 | Not recorded | 6982d281a8b01e76cc8e9267 |
| Random@80% | 80 | Not recorded | 6982d281a8b01e76cc8e9267 |
| Random@60% | 60 | Not recorded | 6982d281a8b01e76cc8e9267 |
| Random@40% | 40 | Not recorded | 6982d281a8b01e76cc8e9267 |
| Random@20% | 20 | Not recorded | 6982d281a8b01e76cc8e9267 |
| Router(Ours) | 9.8 | Not recorded | 6982d281a8b01e76cc8e9267 |
| GPTQ-Int4 | 0 | Not recorded | 6982d281a8b01e76cc8e9267 |

### BF16

High-Precision Ratio: 100. Result ID: 699609f25698665ec441dc9f.

[Reporting paper](<https://www.sota2.com/research/paper/6982d281a8b01e76cc8e9267.md>)

Recorded evaluation configuration: {"Precision":"BF16"}. This record does not assert independent reproduction.

### Random@80%

High-Precision Ratio: 80. Result ID: 699609f25698665ec441dca4.

[Reporting paper](<https://www.sota2.com/research/paper/6982d281a8b01e76cc8e9267.md>)

Recorded evaluation configuration: {"Sampling Ratio":"80%"}. This record does not assert independent reproduction.

### Random@60%

High-Precision Ratio: 60. Result ID: 699609f25698665ec441dca3.

[Reporting paper](<https://www.sota2.com/research/paper/6982d281a8b01e76cc8e9267.md>)

Recorded evaluation configuration: {"Sampling Ratio":"60%"}. This record does not assert independent reproduction.

### Random@40%

High-Precision Ratio: 40. Result ID: 699609f25698665ec441dca2.

[Reporting paper](<https://www.sota2.com/research/paper/6982d281a8b01e76cc8e9267.md>)

Recorded evaluation configuration: {"Sampling Ratio":"40%"}. This record does not assert independent reproduction.

### Random@20%

High-Precision Ratio: 20. Result ID: 699609f25698665ec441dca1.

[Reporting paper](<https://www.sota2.com/research/paper/6982d281a8b01e76cc8e9267.md>)

Recorded evaluation configuration: {"Sampling Ratio":"20%"}. This record does not assert independent reproduction.

### Router(Ours)

High-Precision Ratio: 9.8. Result ID: 699609f25698665ec441dca5.

[Reporting paper](<https://www.sota2.com/research/paper/6982d281a8b01e76cc8e9267.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6982d281a8b01e76cc8e9267.md>)

Recorded evaluation configuration: {"Routing":"Dynamic Mix Precision"}. This record does not assert independent reproduction.

### GPTQ-Int4

High-Precision Ratio: 0. Result ID: 699609f25698665ec441dca0.

[Reporting paper](<https://www.sota2.com/research/paper/6982d281a8b01e76cc8e9267.md>)

Recorded evaluation configuration: {"Quantization":"4-bit"}. This record does not assert independent reproduction.

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