Topic analysis

Orthrus-Qwen3: up to 7.8×tokens/forward on Qwen3, identical output distribution

Orthrus is a dual-architecture framework built on the Qwen3 backbone that unifies the exact generation fidelity of autoregressive large language models (LLMs) with the high-speed parallel token generation of diffusion models. It delivers lossless output, up to 7.8× tokens per forward pass, a ~6x speedup over the Qwen3-8B baseline, higher token acceptance rates, and faster inference than competing methods like EAGLE-3, DFlash, and Fast-dLLM-v2 while avoiding redundant memory overhead from draft models; native integration with vLLM and SGLang is upcoming.