Show HN: Needle: We Distilled Gemini Tool Calling into a 26M Model

The article discusses Needle, a 26M parameter function-calling model that runs on consumer devices. It was open-sourced by Cactus, a company working on an inference engine for mobile and wearables. Needle uses a simple attention network and outperforms other models on single-shot function calling tasks. The model's architecture and weights are available on GitHub and Hugging Face, respectively.

Source →
FeedLens — Signal over noise Last 7 days