LLMLINGUA-2 / LOCAL CPU

Less text. Instructions intact.

Compress background text while keeping your critical instructions exactly as entered.

Paste every critical negation, rule, constraint, or exact evidence here. This field is prepended unchanged to the final output.

Higher keeps more text. The target is approximate; 100% leaves all text unchanged. No rate guarantees meaning is preserved.

—Original tokens
—Final tokens
—Tokens saved
—Reduction

Counts include protected instructions and the separator in both versions. Estimates use LLMLingua’s OpenAI tokenizer; other models may count differently.