February 19th.MediaTekMediaTek Research has now released two lightweight multimodal models with support for Traditional Chinese, the Llama-Breeze2-3B model, which is claimed to work on cell phones, and the Llama-Breeze2-8B model for thin and light laptops.
1AI was informed thatThe series is based on the Meta Llama 3.2 language model.It also supports multimodal input and function calls, and can recognize images and call external tools.
In terms of Traditional Chinese processing capability, the comparison provided by MediaTek shows that compared to the Llama 3.2 3B Instruct model, which has the same number of parameters, the Llama-Breeze2-3B is able to accurately list local famous night markets such as Shihlin Night Market, Raohe Street Night Market, and Luodong Night Market, while the Llama 3.2 3B Instruct model only correctly mentions Shihlin Night Market and also generates two non-existent night markets when composing a short article on night markets in Taipei. Instruct model only mentions Shilin Night Market correctly and generates two non-existent night markets.
In addition, MediaTek has also developed an Android AI Assistant App based on Llama-Breeze2-3B, and at the same time launched an AI text-to-speech model, BreezyVoice, which claims to be able to generate realistic speech in real time with just 5 seconds of sample audio input.
