Training a 210M text-to-image DiT from scratch on one GPU: what I measured [P]

Training large text-to-image DiT model on single GPU shows feasibility and efficiency.

本摘要由 AI 生成，请从原始来源核实细节。