LLM Post-Training: NVIDIA Research and Methods
This Berkeley lecture presentation by NVIDIA Sr. Director of Applied Research Oleksii Kuchaiev covers the landscape of LLM post-training, including data synthesis, mid-training, evaluation methods, reinforcement learning for reasoning and agents, and distillation. The presentation outlines the evolution of post-training into a major computational discipline where the traditional boundary between pre-training and post-training is actively dissolving. Open article →