GGUF conversion of official Nemotron-3-Super-120B-A12B-NVFP4 from NVIDIA using latest mainline llama.cpp (convert_hf_to_gguf.py)

Used llama.cpp b9115. This GGUF doesn't require any forks and works with latest llama.cpp.

Includes two files - with BF16 or Q8_0 quants for some non-expert weights. All expert weights are NVFP4.

These models are much slower than Q4_K GGUFs when offloading experts to the CPU, this is intended for people using Nvidia GPUs with enough VRAM for the whole model and native NVFP4 support.