Wang et al.: MiniLM self-attention relation distillation

What changed

Compressed pretrained Transformers by distilling attention distributions and value-vector relations from the teacher's self-attention module into a smaller student.

Scientific scope

Compressed pretrained Transformers by distilling attention distributions and value-vector relations from the teacher's self-attention module into a smaller student. Knowledge distillation and teacher assistants are inherited; the specific attention/value relation targets are the contribution. Reported retention ratios are task/setup-specific, and later MiniLMv2 or sentence-embedding adaptations are excluded.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Knowledge distillation and teacher assistants are inherited; the specific attention/value relation targets are the contribution. Reported retention ratios are task/setup-specific, and later MiniLMv2 or sentence-embedding adaptations are excluded.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• MiniLM: Deep Self-Attention Distillation for Task-Agnostic Compression of Pre-Trained Transformers
• Wang et al.: MiniLM self-attention relation distillation — official implementation

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction