Text Generation
Transformers
Safetensors
PyTorch
nemotron_h
nvidia
conversational
custom_code
Eval Results

NVFP4 support

#9
by Qnibbles - opened

Dear NVIDIA, can you please release an official NVFP4 version? There is a severe lack of NVFP4 models, and that would help Blackwell hardware really shine.

This article states that the Super and Ultra variants, which are pretrained on NVFP4, are expected to be released in the first half of 2026.
https://developer.nvidia.com/blog/inside-nvidia-nemotron-3-techniques-tools-and-data-that-make-it-efficient-and-accurate/

This article states that the Super and Ultra variants, which are pretrained on NVFP4, are expected to be released in the first half of 2026.
https://developer.nvidia.com/blog/inside-nvidia-nemotron-3-techniques-tools-and-data-that-make-it-efficient-and-accurate/

I hope they are also permissively licensed. Getting tired of models being marketed as open source but really theyre just funnels to the DGX/NIMs platform.

NVIDIA org

Hi @Qnibbles , thanks for your interenst.
The NVFP4 version is released today: https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-NVFP4

@bkartal I notice the officially supported vLLM containers do not yet support the sm12.1 required for NVFP4 on DGX Spark. Do you know if support is imminent or planned?

Hi @Qnibbles , thanks for your interenst.
The NVFP4 version is released today: https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-NVFP4

Very cool, and thanks for letting me know!

bkartal changed discussion status to closed

Sign up or log in to comment