Share your thoughts, 1 month free Claude Pro on usSee more

Commonsense Reasoning on CSQA OOD (test)

82.1Accuracy

MultiRole-R1

Updated 1mo ago

Evaluation Results

Method	Links
MultiRole-R1 2025.07		82.1	87.93
MultiRole-R1 2025.07		82	91.61
SelfConsis SFT 2025.07		81.5	77.6
More think 2025.07		81.2	73.07
SelfConsis SFT+GRPO 2025.07		81.19	89.64
SelfConsis SFT+GRPO 2025.07		81.19	86.4
SelfConsis SFT 2025.07		81	73.5
Zero-shot CoT 2025.07		80.81	57.84
SelfConsis SFT+DPO 2025.07		80.45	64.87
SelfConsis SFT+DPO 2025.07		79.82	69.41
More think 2025.07		79.36	81.77
Self-Refine 2025.07		76.55	82.31
Zero-shot CoT 2025.07		75.85	83.09
Role-Play 2025.07		75.71	79.05
MultiRole-R1 2025.07		75.12	92.98
SelfConsis SFT+GRPO 2025.07		73.71	87.96
Role-Play 2025.07		72.98	48.17
SelfConsis SFT+DPO 2025.07		71.28	82.71
SelfConsis SFT 2025.07		70.86	83.88
More think 2025.07		70.42	84.73
SelfConsis SFT+GRPO 2025.07		69.43	86.85
Self-Refine 2025.07		69.22	50.99
Role-Play 2025.07		67.32	82.27
SelfConsis SFT+DPO 2025.07		67.24	69.83
Zero-shot CoT 2025.07		67.21	83.39
MultiRole-R1 2025.07		66.94	87.96
SelfConsis SFT 2025.07		66.88	83.1
More think 2025.07		64.5	85.85
Self-Refine 2025.07		63.77	82.61
Zero-shot CoT 2025.07		63.06	83.83
Role-Play 2025.07		55.07	76.2
Self-Refine 2025.07		54.02	77.61