Clover-35B-A3B-Preview

Repo: DiffuSpace/Clover-35B-A3B-Preview Weights and the decoding settings: huggingface.co Launch post with the speed numbers: diffuspace.ai

#DiffuSpace has introduced Clover-35B-A3B-Preview. Built on Qwen3.6-35B-A3B, #Clover combines a text-only mixture-of-experts backbone with block diffusion generation: DiffuSpace/Clover-35B-A3B-Preview t.co

🔗 Get started with Clover-35B-A3B-Preview: chat.diffuspace.ai • HuggingFace: huggingface.co • GitHub: DiffuSpace/Clover-35B-A3B-Preview • ModelScope: modelscope.cn

Diffuspace/Clover-35B-A3B-Preview · Hugging Face Clover-35B-A3B-Preview is the first model released by DiffuSpace, delivering state-of-the-art performance among open-source diffusion language models at release on the benchmarks reported in our launch blog. Clover combines a 35B-parameter mixture-of-experts backbone with approximately 3B active parameters per token, block diffusion generation, and a 128K context window. Blog · Code & inference backends · Hugging Face

官方博客：diffuspace.ai Hugging Face（模型卡写了 derived from Qwen3.6-35B-A3B）：huggingface.co GitHub：DiffuSpace/Clover-35B-A3B-Preview

DiffuSpace's Clover-35B-A3B-Preview (open-weight) An open-weight production diffusion language model (dLLM) with 35B total parameters capable of 680 tps on a single card at BF16. Blog: diffuspace.ai Repo: DiffuSpace/Clover-35B-A3B-Preview Model: huggingface.co

Code for using Clover (600+ tps) for self-speculative decoding with SGLang and llama.cpp is available at DiffuSpace/Clover-35B-A3B-Preview