# Multi-Turn Consistency on Role-playing dataset

Source: [Multi-Turn Consistency on Role-playing dataset](<https://www.sota2.com/research/sota/multi-turn-consistency-on-role-playing-dataset>)

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

Data updated: 2026-05-19T12:42:56.814Z

Evaluation of model performance on multi-turn conversational consistency using the Role-playing dataset, measured by KE, KA, and PU metrics.

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-Turn Consistency |
| Dataset (including any stated split) | Role-playing dataset |
| Metric | KE Score |
| Unit | SCALAR |
| Direction | Higher is better |
| Results loaded | 12 of 12 |
| 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

- [KE Score](<https://www.sota2.com/research/sota/multi-turn-consistency-on-role-playing-dataset.md?metric=KE%20Score>): Higher is better; unit: SCALAR.

- [KA Score](<https://www.sota2.com/research/sota/multi-turn-consistency-on-role-playing-dataset.md?metric=KA%20Score>): Higher is better; unit: SCALAR.

- [PU Score](<https://www.sota2.com/research/sota/multi-turn-consistency-on-role-playing-dataset.md?metric=PU%20Score>): Higher is better; unit: SCALAR.

- [Average Score](<https://www.sota2.com/research/sota/multi-turn-consistency-on-role-playing-dataset.md?metric=Average%20Score>): Higher is better; unit: SCALAR.

## Recorded results

| Method | KE Score | Result date | Report paper |
| --- | --- | --- | --- |
| PPR-GDE | 2.65 | Not recorded | 6a0bef08dbc836b87d82c9ae |
| GRPO | 2.61 | Not recorded | 6a0bef08dbc836b87d82c9ae |
| PPR-GDE | 2.57 | Not recorded | 6a0bef08dbc836b87d82c9ae |
| PPO | 2.48 | Not recorded | 6a0bef08dbc836b87d82c9ae |
| GRPO | 2.45 | Not recorded | 6a0bef08dbc836b87d82c9ae |
| PPO | 2.42 | Not recorded | 6a0bef08dbc836b87d82c9ae |
| PPR-GDE | 2.39 | Not recorded | 6a0bef08dbc836b87d82c9ae |
| PPO | 2.37 | Not recorded | 6a0bef08dbc836b87d82c9ae |
| GRPO | 2.33 | Not recorded | 6a0bef08dbc836b87d82c9ae |
| w/o RL | 2.23 | Not recorded | 6a0bef08dbc836b87d82c9ae |
| w/o RL | 2.17 | Not recorded | 6a0bef08dbc836b87d82c9ae |
| w/o RL | 1.82 | Not recorded | 6a0bef08dbc836b87d82c9ae |

### PPR-GDE

KE Score: 2.65. Result ID: 6a0c5aaefbc373592c03e1c8.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bef08dbc836b87d82c9ae.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bef08dbc836b87d82c9ae.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-3B-Instruct"}. This record does not assert independent reproduction.

### GRPO

KE Score: 2.61. Result ID: 6a0c5aaefbc373592c03e1c7.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bef08dbc836b87d82c9ae.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73101e209131cea5a395.md>)

[Implementation](<https://github.com/deepseek-ai/DeepSeek-Math>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-3B-Instruct"}. This record does not assert independent reproduction.

### PPR-GDE

KE Score: 2.57. Result ID: 6a0c5aaefbc373592c03e1cc.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bef08dbc836b87d82c9ae.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bef08dbc836b87d82c9ae.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-1.5B-Instruct"}. This record does not assert independent reproduction.

### PPO

KE Score: 2.48. Result ID: 6a0c5aaefbc373592c03e1c6.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bef08dbc836b87d82c9ae.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa63a01e209131ce9ffdaa.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-3B-Instruct"}. This record does not assert independent reproduction.

### GRPO

KE Score: 2.45. Result ID: 6a0c5aaefbc373592c03e1cf.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bef08dbc836b87d82c9ae.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73101e209131cea5a395.md>)

[Implementation](<https://github.com/deepseek-ai/DeepSeek-Math>)

Recorded evaluation configuration: {"Backbone":"Qwen3-0.6B"}. This record does not assert independent reproduction.

### PPO

KE Score: 2.42. Result ID: 6a0c5aaefbc373592c03e1ce.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bef08dbc836b87d82c9ae.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa63a01e209131ce9ffdaa.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-0.6B"}. This record does not assert independent reproduction.

### PPR-GDE

KE Score: 2.39. Result ID: 6a0c5aaefbc373592c03e1d0.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bef08dbc836b87d82c9ae.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bef08dbc836b87d82c9ae.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-0.6B"}. This record does not assert independent reproduction.

### PPO

KE Score: 2.37. Result ID: 6a0c5aaefbc373592c03e1ca.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bef08dbc836b87d82c9ae.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa63a01e209131ce9ffdaa.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-1.5B-Instruct"}. This record does not assert independent reproduction.

### GRPO

KE Score: 2.33. Result ID: 6a0c5aaefbc373592c03e1cb.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bef08dbc836b87d82c9ae.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73101e209131cea5a395.md>)

[Implementation](<https://github.com/deepseek-ai/DeepSeek-Math>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-1.5B-Instruct"}. This record does not assert independent reproduction.

### w/o RL

KE Score: 2.23. Result ID: 6a0c5aaefbc373592c03e1cd.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bef08dbc836b87d82c9ae.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-0.6B"}. This record does not assert independent reproduction.

### w/o RL

KE Score: 2.17. Result ID: 6a0c5aaefbc373592c03e1c5.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bef08dbc836b87d82c9ae.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-3B-Instruct"}. This record does not assert independent reproduction.

### w/o RL

KE Score: 1.82. Result ID: 6a0c5aaefbc373592c03e1c9.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bef08dbc836b87d82c9ae.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-1.5B-Instruct"}. This record does not assert independent reproduction.

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