Hacker News Reader

Jbungaon Show HN: Reducing LLM input tokens by 70%

Fair criticism. I replaced that borrowed hero treatment and redeployed the homepage with a native Rose/Adola hero today. It was a bad first impression for a technical audience, especially for something asking people to trust an API in their LLM pipeline.

Jbungaon Show HN: Reducing LLM input tokens by 70%

Yeah, you can set roughly the target ratio through the api (for example, target_ratio=.3), though our api will try to maximize the quality given the this target ratio (and it might add a couple more tokens to do so)

Jbungaon Adola: Reducing LLM input tokens by 70%

Most input tokens are kind of useless lol, so here we are