Hunyuan Image2.0
Search documents
MiniMax最新语音大模型超越OpenAI,取得国际评测榜单第一;Meta发布CATransformers框架丨AIGC日报
创业邦· 2025-05-17 00:55
Group 1 - MiniMax's new speech model Speech-02 surpasses OpenAI and ElevenLabs, achieving top rankings in two international voice evaluation lists, with state-of-the-art results in core technical indicators like Word Error Rate (WER) and Similarity Index (SIM) [1] - The United States and the UAE are collaborating to establish the largest artificial intelligence data center outside the US, with a capacity of 5GW and covering approximately 26 square kilometers, supported by the Abu Dhabi government and G42 [2] - Tencent has launched the Hunyuan Image2.0 model, which features a tenfold increase in parameter scale and supports multiple interaction methods including text, voice, and sketches [3] Group 2 - Meta's FAIR research team, in collaboration with Georgia Tech, introduced the CATransformers framework, which utilizes multi-objective Bayesian optimization to evaluate model architecture and hardware performance, aiming to balance latency, energy consumption, accuracy, and carbon footprint [4] - CATransformers specifically targets edge inference devices, producing variants of large CLIP models that reduce carbon emissions by 17% while maintaining performance, with CarbonCLIP-XS achieving an 8% accuracy improvement and lower carbon emissions [2]