lora_moe_merged

run main.py, then merge.py. Runs on 24GB vram