chiennv2000/orthrus

[NeurIPS 2026 Spotlight] Fast, lossless LLM inference via dual-view diffusion decoding.

246 best points · 45 comments · 1 Hacker News post · 483 GitHub stars · View on GitHub

Hacker News discussions

• Orthrus-Qwen3: up to 7.8×tokens/forward on Qwen3, identical output distribution (246 points, 45 comments, by FranckDernoncou, 2026-05-15)