EngineersofAI LogoCoursesBlogResearch LabAI LettersThe LabCode BankInteractive 3DKodrEarnestJobs
Skip to main content
Interactive 3D/Context Compression for Long Inputs
Controls
Strategy
Compression50%
10%90%
Query
Stats
Original tokens735
Compressed to-
Info retention-
Latency saving-
Relevance Score
>0.8 High
0.6–0.8 Medium
0.4–0.6 Low
<0.4 Drop

Context Compression for Long Inputs - Interactive Visualization

When documents exceed the context window, compression is necessary. Extractive compression keeps the most relevant sentences (fast, lossless for selected content). Abstractive compression summarizes (slower, more concise). Token-level pruning (LLMLingua) removes the least informative tokens while preserving syntax. This demo compares all three on the same document at adjustable compression ratios.

  • Extractive compression: top-K sentences selected by relevance score, rest discarded
  • Abstractive compression: document summarized into a shorter paraphrase losing some detail
  • LLMLingua token pruning: per-token perplexity scored, low-information tokens dropped
  • Compression ratio slider: set target ratio and see token count, quality score, and latency

Part of the EngineersOfAI Interactive 3D - free interactive visualizations covering every major concept in machine learning and AI engineering. Hover any element for a plain-English explanation. No code required.