dparallel-diffusion-llm-parallel-decoding

【部署与运维】Enable Diffusion Language Models to achieve 8.5x inference speedup (24-30 steps vs. 256) through certainty-forcing distillation that trains models to achieve simultaneous high confidence across multiple tokens. Use when optimizing inference latency for dLLM deployments.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准