Zero-shot Reasoning on (PIQA, HellaSwag, ARC-e, ARC-c, OBQA, Race, WSC273, LAMBADA, MMLU)

78.07PIQALLaMA-2-7B-hf

Evaluation results

Method
LLaMA-2-7B-hf2026.02	78.07	76.02	76.3	46.25	44.2	39.62	80.59	73.88	41.77	61.86
Surrogate-Assisted Layer Contribution Estimation2026.02	77.09	71.85	71.55	40.96	39.4	36.94	79.12	66.82	36.63	57.82
SLEB2026.02	76.44	71.8	71.38	39.85	40.8	35.5	78.39	62.16	30.84	56.35
ShortGPT2026.02	75.73	71.62	71.04	42.92	40.4	38.47	76.92	61.67	35.1	57.09
Shortened-LLaMA2026.02	74.97	72.98	72.01	43.6	40.4	37.99	82.78	63.01	35.72	58.16
Surrogate-Assisted Layer Contribution Estimation2026.02	73.72	67.19	64.73	37.29	38.6	36.46	75.46	47.39	29.55	52.27
SLEB2026.02	73.01	66.56	67	39.51	38.8	35.02	73.99	44.77	23.78	51.38
Shortened-LLaMA2026.02	70.95	65.28	59.34	37.97	37.4	33.3	76.92	47.47	25.89	50.5
ShortGPT2026.02	70.95	65.28	59.34	37.97	37.4	33.3	76.92	47.47	25.89	50.5
Surrogate-Assisted Layer Contribution Estimation2026.02	70.46	58.4	58.21	34.04	36	33.01	70.33	34.27	27.25	46.89
SLEB2026.02	69.1	56.4	59.47	32.51	35.2	32.63	68.5	31.34	23.72	45.43
SliceGPT2026.02	66.76	52.99	46.63	30.2	31.4	33.97	80.22	31.52	25	44.3
Surrogate-Assisted Layer Contribution Estimation2026.02	66.59	46.18	49.37	28.76	32.6	29.89	61.17	17.31	23.7	39.51
Shortened-LLaMA2026.02	64.85	56.17	48.02	32.76	32.8	32.25	71.43	29.15	38.11	45.06
ShortGPT2026.02	64.85	56.17	48.02	32.76	32.8	32.25	71.43	29.15	38.11	45.06
SLEB2026.02	61.86	46.65	44.91	30.2	31	30.24	57.88	15.27	25.07	38.12
SliceGPT2026.02	60.77	42.7	35.9	27.56	27	30.81	74.73	23.95	24.41	38.65
Shortened-LLaMA2026.02	60.23	44.3	36.11	30.63	27.6	29.09	60.81	5.05	33.73	36.4
ShortGPT2026.02	59.52	43.71	41.58	30.03	35	29	68.86	11	33.32	39.11
SliceGPT2026.02	58.87	38.56	32.45	26.19	27	29.47	72.53	18.46	24.52	36.45
SliceGPT2026.02	57.18	34.75	30.77	25	25.6	27.75	69.23	14.5	24.93	34.41