MiniLMv2: Multi-Head Self-Attention Relation Distillation for Compressing Pretrained Transformers
read the original abstract
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.
This paper has not been read by Pith yet.
Forward citations
Cited by 5 Pith papers
-
Contribution Weights: A Geometrical Analysis of Self-Attention Transformers
Contribution Weights combine attention, value magnitude, and directional alignment to measure token influence more faithfully than attention alone, and show attention sinks actively suppress information via a convex s...
-
DeepSurvey: Enhancing Analytical Depth and Citation Reliability in Automated Survey Generation
DeepSurvey introduces an agentic system for automated survey generation that improves depth through full-text keynotes, cross-paper clustering, and code analysis, while boosting citation reliability via graph expansio...
-
AtlasKV: Augmenting LLMs with Billion-Scale Knowledge Graphs in 20GB VRAM
AtlasKV integrates billion-scale KGs into LLMs parametrically with sub-linear complexity and low memory by converting triples into key-value representations handled by the model's attention.
-
DeBERTaV3: Improving DeBERTa using ELECTRA-Style Pre-Training with Gradient-Disentangled Embedding Sharing
DeBERTaV3 improves DeBERTa by switching to replaced token detection pre-training and using gradient-disentangled embedding sharing, reaching 91.37% on GLUE and new SOTA on XNLI zero-shot.
-
Structural Pruning of Large Vision Language Models: A Comprehensive Study on Pruning Dynamics, Recovery, and Data Efficiency
Widthwise pruning of LVLM language backbones combined with supervised finetuning and hidden-state distillation recovers over 95% performance using just 5% of data across 3B-7B models.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.