Downstream Task Evaluation on LAMBADA, ARC, WinoGrande, PIQA, HellaSwag, SciQ, and RACE benchmark leaderboard · SOTA2 Research

Downstream Task Evaluation on LAMBADA, ARC, WinoGrande, PIQA, HellaSwag, SciQ, and RACE

45LAMBADA (OpenAI)PonderLM-2

Evaluation Results

Method	Links
PonderLM-22026.03		45	47.6	36.6	21.9	51.6	64.4	32.3	80.6	31.4	45.7
PonderLM-32026.03		43.3	48.7	35	21.9	50.7	64.2	32.2	80	29.3	45
PonderLM-32026.03		40.8	50.7	36	22	52.6	64.9	32.3	86.9	31.8	46.4
LoopedLM2026.03		40.4	47.2	28.6	20.4	50.8	64.4	30.6	75.4	28.8	43
PonderLM-12026.03		39.6	47.8	29.3	20.4	50	64.1	30.4	77.2	29.9	43.2
Pause2026.03		38.1	47.5	27.8	20.3	50.3	63.8	29.8	76.3	29	42.5
PonderLM-22026.03		37.6	50.6	34	21.8	53.5	65	32.5	87.5	31.2	46
MoR2026.03		37	45.3	26.6	19.9	50.1	63.8	30	77.3	28.6	42
LoopedLM2026.03		34.2	48.2	26.6	20.6	47	64.2	30.7	83.4	27.8	42.5
PonderLM-12026.03		34	49.1	28.4	21.8	49.7	64.1	30.9	84.4	29.5	43.5
MoR2026.03		31.5	48.5	25.3	20	51	63.7	30.6	84.7	28	42.6
Pause2026.03		30.7	46.9	26.3	21.8	51.6	63.4	30.1	81.8	28.5	42.3