Natural Language Understanding and Reasoning on Standard Downstream Benchmarks Two-Shot (val) benchmark leaderboard · SOTA2 Research

Natural Language Understanding and Reasoning on Standard Downstream Benchmarks Two-Shot (val)

56.86ARC-E Accuracy (Normalized)AdaGC

Evaluation Results

Method	Links
AdaGC2025.02		56.86	29.61	59.36	57.89	33.6	73.99	57.62	26.46	85.9	53.47
GlobalGC2025.02		55.81	28.58	60.7	56.54	33	73.72	56.75	25.51	83.2	52.64
AdaGC2025.02		53.83	28.42	58.69	55.66	33.8	73.07	54.14	25.12	81.8	51.61
AGC2025.02		52.95	28.67	56.15	55.69	35.4	73.07	56.43	26.88	82.8	52
Clippy2025.02		52.86	29.1	56.48	53.76	31.8	73.07	55.72	26.03	82.6	51.27
ClipByValue2025.02		51.94	26.88	57.55	53.36	32.4	72.31	54.14	26.63	81.6	50.75
GlobalGC2025.02		50.34	27.39	58.81	52.96	34.2	71.16	54.06	25.37	79.9	50.47
GlobalGC2025.02		47.26	25.6	50.31	46.44	32.2	69.64	52.33	25.07	77.8	47.41
ClipByValue2025.02		47.1	25.77	56.54	43.97	30	68.88	52.96	26.09	77.2	47.61
Clippy2025.02		46.55	25.85	49.76	45.71	30	70.02	53.2	25.69	77.7	47.16
AdaGC2025.02		46.04	26.19	49.72	47.51	31	69.7	54.38	24.98	78.5	47.56