Hugging Face researchers present methods to reduce GPU memory in LLM training, enabling single-GPU distillation and lowering costs for large models.
Maximizing AI Efficiency: Affordable Knowledge Distillation Strategies For Big Data
