Are there any open-weights diffusion LLM models I can play with on my local hardware? Curious about the performance delta of this style of model in more resource constrained scenarios (i.e. consumer Nvidia GPU, not H100s in the datacenter)