This involves the following improvements:

• Add Gemma 27B -> https://huggingface.co/google/gemma-3-27b-it This involves checking its compatibility with executing alongside Llama 8B and adding support to use the "image" prompts.
• [STRETCH] Add Qwen3 14B -> https://huggingface.co/Qwen/Qwen3-14B-FP8
• [STRETCH] Add Qwen3 30B A3B -> https://huggingface.co/Qwen/Qwen3-30B-A3B-FP8

Both Qwen3-14B and Qwen3-30B will not be hosted immediately.