Key Responsibilities Focus on the training and inference optimization of foundation models (LLM/VLM), low-bit quantization techniques, operator fusion and kernel-level optimization, as well as the research and development of the next generation of Spatial Foundation Models.
LLM/VLM Acceleration and Optimization - Focus on efficient training and inference acceleration of LLM/VLM models, including but not limited to mixed-precision training, KV cache management, and the deployment of low-bit quantization techniques;
- Continuously track the evolution of basic model operators (such as flash-attention) and emerging algorithmic architectures and computational paradigms.
Spatial Foundation Model Algorithm R&D - Participate in the development of spatial foundation models, focusing on 3D scene generation and reconstruction, 3D representation learning, semantic understanding, including but not limited to 3D VAE, 3D Points Cloud ...