Topic Classification and Text Generation on AGNews (test) benchmark leaderboard · SOTA2 Research

Topic Classification and Text Generation on AGNews (test)

16.94PPL (Output)GPT2(raw)

Evaluation Results

Method	Links
GPT2(raw)2022.12		16.94	74.41	55.75	52.17	83.28	46.88	45.55
PF2022.12		20.27	32.35	68.67	68.44	87.14	59.17	32.73
Ctr-PF2022.12		20.41	33.9	83.25	83.21	95.47	60.34	31.2
GPT22022.12		22.22	23.46	82.92	83.08	95.23	54.93	39.93
+PT2022.12		23.74	27.88	83.5	83.55	95.49	57.89	36.02
+PT+noise2022.12		26.39	27.02	82.42	82.45	94.58	58.06	35.53
+PT(noise)+PL2022.12		30.62	13.96	87.83	87.48	97.42	47.11	56.67
Test set2022.12		31.04	—	94.88	94.89	99.34	67.24	23.31
+PT(select)+PL2022.12		31.34	14.07	87.92	87.54	97.46	46.71	57.33
+PT(noise)+PL2022.12		32.36	16.64	89.67	89.7	98.11	53.79	47.95
+PT(select)+PL2022.12		33.23	16.66	90.5	90.52	98.31	53.71	47.69
DuNST2022.12		34.73	33.58	93.58	93.59	98.99	59.42	37.02
T52022.12		48.33	32.12	88.33	88.43	97.95	58.06	37.01
UniLM2022.12		55.79	36.28	87.67	87.7	96.3	54.76	43.77
+PT2022.12		57.4	40.95	86.42	86.36	96.69	52.35	46.41
+PT+noise2022.12		58.59	45.32	85.42	85.27	95.88	53.35	46.57