Dumbify

• are your prompts too long for an llm with smaller context window
• is your budget too tight for a llm with a greater context window
• Dont Worry the Dumbify fixes this issue using prompt compressions
• “but won’t it cause context loss ?” great questions and answer and solution to which is given by a context extraction from previous chat history.
• This is realized using a fine-tuned microsoft/LLMLingua-2 117M parameters for prompt compression and fine-tuned google/gemma-3-270m-it for context extraction.
• total models weighting just over 1GiB making this entire setup cloud friendly.
• This entire setup takes requests from you and gets you the response, so no juggeling between API’s to get what you want.