compression

Definition

Representing information — or a model — in fewer bits while preserving what matters. Practically it covers quantization, pruning, and distillation, all aimed at cutting memory and latency for deployment. Theoretically it runs deeper: good prediction and good compression are formally linked, so some papers evaluate language models by how well they compress held-out text rather than by task accuracy.

Explainers using this term

• Paper Explained: DeepSeek-V4.1-Flash — How Far Can KV Cache Compression Go?DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression
• Paper Walkthrough: X-AuT — Pull Audio-Encoder Layers Out Two at a Time, Then Backfill from a Bigger TeacherX-AuT: Progressive Audio-Encoder Compression for Speech LLMs with Cross-Scale Distillation
• Paper walkthrough: BeaconKV — evicting KV cache by predicting when a reasoning model will re-read its own planBeaconKV: Key-Value Cache Compression Guided by Beacon Queries for Efficient Large Reasoning Model Inference