Prefix-Tuning for Arbitrary Output Sequences on Pretrained Transformers · Attention Matching

Peter Cho-Ho Lam¹, Ziyi Wang¹, Zirui Zhou¹

¹Huawei Technologies Canada

1

2

		k = 1	k = 5	k = 10
GPT-2		98.2%	91.4%	79.3%
Qwen3-4B		100%	100%	100%