Adding support for LoRa and other dllm like Llada and Fast dllm

[ayushnangia]

• Could you add Adapter-based finetuning dramatically lowers the memory/barrier to customizing dLLMs?
• Fast-dLLM and future dLLMs (Dream, PRISM, etc.) follow a similar block-diffusion recipe; adding supporting for them through ModelRegistry would be great!
  https://github.com/NVlabs/Fast-dLLM/tree/main/v2
  https://github.com/ML-GSAI/LLaDA

[jzhao2024]

we'd release this support soonish.
prolly upon the full release of llada-2.0 this month :)
thanks for the interests. stay tuned.