chiennv2000/orthrus
Dual-architecture framework combining autoregressive LLM fidelity with parallel diffusion token generation for fast, lossless inference.
Not currently ranked — collecting fresh signals.
star history
Orthrus implements a dual-view diffusion decoding approach that unifies autoregressive generation accuracy with diffusion model parallelization. The framework modifies Qwen3 with specialized components enabling 4-5x inference speedups while guaranteeing strictly lossless output compared to the base model. Available in 1.7B, 4B, and 8B parameter variants.
Frequently asked
- What is chiennv2000/orthrus?
- Dual-architecture framework combining autoregressive LLM fidelity with parallel diffusion token generation for fast, lossless inference.
- Is orthrus open source?
- Yes — chiennv2000/orthrus is open source, released under the MIT license.
- What language is orthrus written in?
- chiennv2000/orthrus is primarily written in Python.
- How popular is orthrus?
- chiennv2000/orthrus has 457 stars on GitHub.
- Where can I find orthrus?
- chiennv2000/orthrus is on GitHub at https://github.com/chiennv2000/orthrus.