Lossless Prompt Compression via Dictionary-Encoding and In-Context Learning: Enabling Cost-Effective LLM Analysis of Repetitive Data

Implement a library for 'dictionary-based prompt compression' where you replace frequent strings with tokens for lower latency and cost. Ensure it works seamlessly with existing tokenizers.

Suggested repo: dictzip-llm

"Compress your context window by 5x without losing a single token of info."

Estimated effort: 50h