Papers › MiniLMv2: Multi-Head Self-Attention Relation Distillation for Compressing Pretrained...

MiniLMv2: Multi-Head Self-Attention Relation Distillation for Compressing Pretrained Transformers

31 Dec 2020Findings (ACL) 2021 8arXiv:2012.15828archive 2025-07-28

Wenhui Wang, Hangbo Bao, Shaohan Huang, Li Dong, Furu Wei

We generalize deep self-attention distillation in MiniLM (Wang et al., 2020) by only using self-attention relation distillation for task-agnostic compression of pretrained Transformers. In particular, we define multi-head self-attention relations as scaled dot-product between the pairs of query, key, and value vectors within each self-attention module. Then we employ the above relational knowledge to train the student model. Besides its simplicity and unified principle, more favorably, there is no restriction in terms of the number of student's attention heads, while most previous work has to guarantee the same head number between teacher and student. Moreover, the fine-grained self-attention relations tend to fully exploit the interaction knowledge learned by Transformer. In addition, we thoroughly examine the layer selection strategy for teacher models, rather than just relying on the last layer as in MiniLM. We conduct extensive experiments on compressing both monolingual and multilingual pretrained models. Experimental results demonstrate that our models distilled from base-size and large-size teachers (BERT, RoBERTa and XLM-R) outperform the state-of-the-art.

PaperPDFConference PDFCode

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

facebookresearch/data2vec_vision mentioned on GitHubpytorch report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

XLM-R

1 archive task tag without a task page not shown.

Results from the paper archive 2025-07-28

No leaderboard rows for this paper in the archive.

Methods

Absolute Position EncodingsAdamAttentionAttention DropoutBERTBPEDense ConnectionsDropoutLabel SmoothingLayer NormalizationLinear LayerLinear Warmup With Linear DecayMulti-Head AttentionPosition-Wise Feed-Forward LayerResidual ConnectionRoBERTaSoftmaxTransformerWeight DecayWordPiece

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections