Llama-3.3-Nemotron-Super-49B-v1.5

NVIDIA's 49B reasoning/chat model, running 4-bit quantized on ZeroGPU. Reasoning ON shows a <think> trace before the answer.