Framework speeds up diffusion-based language models by nine times

Original: SAID: Accelerating Diffusion-Based Language Models via Scaffold-Aware Iterative Decoding

Source: arxiv.org ↗