1. Core algorithm R&D for Large Language Models (LLMs), including pre-training, supervised fine-tuning (SFT), RLHF, alignment optimization, and inference enhancement.
2. Explore efficient model tuning strategies and high-quality data construction methods; research advanced model architectures such as MoE sparsification and Latent Attention.
3. Support the application and optimization of LLMs across diverse business scenarios, including search, recommendation, dialogue, AIGC, speech, cloud storage, document editing, and overseas products.
4. Design, implement, and optimize large-scale distributed training and inference frameworks to improve training stability and inference efficiency.
5. Contribute to the platformization of LLMs, driving the transformation of model capabilities into innovative products.
