Tencent ships Hy3 preview as an open-weight agent model
- Source
- x.com
- Date
Tencent has launched Hy3 preview, the latest and most capable model in the Hy series. Strong reasoning, long-context understanding, and agent capabilities built for real-world workflows. Now open-source on @github, @huggingface, @ModelScope2022, and GitCode. Available now: https://t.co/3MRBH10GBY #AI #LLM
Context
Tencent's Hy3 preview gives developers a downloadable model they can adapt for their own tasks. Its model card provides deployment instructions and training support, including fine-tuning, which adjusts an existing model using additional examples.
The release uses a design: it contains 295 billion parameters, but activates 21 billion for each text token. That distinction explains how the model can have a large overall capacity while using only part of it for each processing step. The preview is distributed under Tencent's community license.
- mixture-of-experts — A model built from many specialist sub-networks where only a few activate per token, giving big-model capability at small-model running cost.
- open weights — A model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
- tool use — A model's ability to call external functions — run code, search the web, edit files — instead of only generating text.
Checking sign-in…
Loading comments…