Generative Commonsense Inference on ATOMIC 2021 (test) benchmark leaderboard · SOTA2 Research

Generative Commonsense Inference on ATOMIC 2021 (test)

50.71BLEU-1CANDLE Distilled

Evaluation Results

Method	Links
CANDLE Distilled2024.01		50.71	33.85	25.55	20.43	32.45	51.91	69.68	76.86	78.5
GPT2-XL2024.01		45.72	29.18	21.12	16.15	29.97	49.69	64.61	76.09	70.5
GPT2-XL2024.01		45.38	29.2	21.09	16.15	30.09	49.86	65.02	75.89	77.5
GPT2-XL2024.01		45.3	29.08	21	16.06	29.98	48.61	63.98	75.56	71.5
CANDLE Distilled2024.01		43.86	24.4	15.12	10	28.36	43.86	54.25	72.94	79.5
LLAMA22024.01		42.78	23.64	14.58	9.471	27.74	42.55	53.12	71.51	71
GPT2-XL2024.01		42.15	25.77	17.82	13.14	29.82	47.61	63.7	70.39	76.5
LLAMA22024.01		42.07	23.08	14.14	9.198	28.14	42.75	53.69	71.93	76.5
LLAMA22024.01		42.04	23.01	14.1	9.125	27.8	42.9	53.17	71.52	68.5
LLAMA22024.01		41.07	22.46	13.62	8.619	27.74	42.42	53.28	71.77	74
LLAMA22024.01		31.22	16.87	9.767	5.989	19.74	27.67	17.83	58.41	65.5
ChatGPT2024.01		26.32	12.5	7.16	4.415	18.6	24.65	8.313	58.69	81
LLAMA22024.01		18.26	7.453	3.594	1.945	15.9	20.28	8.872	48.23	48.5
ChatGPT2024.01		11.82	4.258	1.891	0.926	13.87	13.73	4.35	49.28	78.5
ChatGPT2024.01		9.906	3.568	1.556	0.736	11.85	11.02	2.905	46.17	64
GPT2-XL2024.01		4.35	1.598	0.732	0.293	5.702	5.03	0.792	37.11	14.5