Hunyuan

About

The Hunyuan large language model (LLM) family, developed by Tencent, is renowned for its resource-efficient design and high performance, utilizing Mixture of Experts (MoE) architectures to activate specialized sub-models dynamically for diverse tasks. Its flagship model, Hunyuan-Large (Hunyuan-MoE-A52B), stands out with 389 billion parameters, of which 52 billion are activated, making it a leading open-source Transformer-based MoE model. It excels in various areas, including language understanding, generation, reasoning, and problem-solving. Tencent offers several tailored versions of Hunyuan-Large, such as instruction-tuned models and those optimized for tasks like text-to-image generation (Hunyuan-DiT). These models are accessible via APIs on Tencent Cloud and Hugging Face, supporting extensive AI community research and application 1513.