Table 2:

Comparison of (micro-F1) performance with recent generative (except SpERT) approaches in RE. Relation triplets/pairs are considered correct only if both of the corresponding entity types are correctly generated.

	Method	Params	CONLL	ADE	NYT
1. Fully supervised	a. SpERT* (Eberts and Ulges, 2019b)	110M	71.54	79.22	−
	b. TANL (Paolini et al., 2021)	220M	71.48	80.61	90.83
	c. TANL (MT) (Paolini et al., 2021)	220M	72.66	80.00	90.52
	d. REBEL (Huguet Cabot and Navigli, 2021)	460M	75.44	82.21	92.00
	e. Flan T5 (Large) (Chung et al., 2022)	760M	75.28	83.15	91.03
	f. + GPT-3-generated CoT	760M	80.76	92.17	95.23
2. Few-shot	a. In-Context GPT-3 (Brown et al., 2020a)	175B	76.53	82.66	61.79
	b. + CoT	175B	78.18	−	−
	c. Flan T5 (Large) w/CoT Explanations and reference labels generated from GPT-3	760M	76.13	−	−