VikramPal/Qwen3-Omni-30B-A3B-Thinker-QLoRA
Adapter vs Qwen/Qwen3-Omni-30B-A3B-Instruct. VikramPal/Qwen3-Omni-30B-A3B-Thinker-QLoRA is a LoRA adapter, not full weights: rank 16, on fc1, fc2, k_proj, o_proj, out_proj, q_proj, v_proj in 48 layers (384 adapted matrices). It modifies Qwen/Qwen3-Omni-30B-A3B-Instruct at load time.