Probability Calibration on ChaosNLI combined SNLI+MNLI K=3 benchmark leaderboard · SOTA2 Research

Probability Calibration on ChaosNLI combined SNLI+MNLI K=3

2.15ECEIR-Soft

Evaluation Results

Method	Links
IR-Soft2026.03		2.15	3.96	7.2	0.554	0.942
Dirichlet-Soft2026.03		2.57	1.71	2.03	0.51	0.839
IR-Soft2026.03		2.65	2.59	6.24	0.549	0.934
LS-TS2026.03		2.65	2.99	6.32	0.529	0.881
SoftPlatt2026.03		2.66	2.32	2.43	0.509	0.839
MCTS S=12026.03		2.82	2.35	4.6	0.519	0.862
SoftPlatt2026.03		3.17	2.92	3.18	0.511	0.841
Dirichlet-Soft2026.03		3.19	2.71	2.95	0.509	0.836
MCTS S=12026.03		3.2	2.98	6.3	0.529	0.877
SLTS2026.03		3.22	2.59	4.62	0.52	0.862
SLTS2026.03		3.45	3.36	6.53	0.529	0.877
VS2026.03		3.46	3.07	5.13	0.518	0.857
Oracle TS2026.03		3.48	2.83	4.64	0.52	0.862
Oracle TS2026.03		3.66	3.54	6.66	0.529	0.876
VS2026.03		3.91	3.35	6.46	0.525	0.869
LS-TS2026.03		4.16	3.72	5.61	0.522	0.873
TS2026.03		10.55	10.38	7.3	0.537	0.886
Platt2026.03		11.16	10.92	7.28	0.53	0.875
ATS2026.03		11.36	11.29	7.73	0.538	0.883
Dirichlet-Hard2026.03		11.55	11.4	7.64	0.535	0.889
TS2026.03		11.63	11.57	8.48	0.549	0.9
Platt2026.03		12.43	12.36	8.43	0.539	0.896
Dirichlet-Hard2026.03		12.69	12.63	8.84	0.539	0.895
ATS2026.03		13.09	13.07	9.06	0.552	0.9
Uncalibrated2026.03		27.79	27.77	18.61	0.65	1.384
Uncalibrated2026.03		35.97	35.92	24.02	0.743	2.148