Tencent's Hy3 Model Features 295 Billion Parameters with 5.4% Hallucination Rate

Tencent has released the Hy3 open-source language model, which boasts a total of 295 billion parameters, with 21 billion active parameters at any given time. The model utilizes a mixture-of-experts architecture, allowing it to match the performance of models two to five times its size while achieving a hallucination rate of just 5.4%, effectively halving the typical rate seen in similar models. This release positions Tencent as a significant player in the open-source AI landscape, following their previous advancements in model architecture. Developers can leverage Hy3 for applications requiring high efficiency and lower error rates, potentially reshaping their approach to model selection and deployment. More details can be found in The Decoder.