Nemotron-3-Nano-Omni#

Nemotron-3-Nano-Omni-30B-A3B-Reasoning is NVIDIA’s omnimodal reasoning model. It pairs a NemotronH (hybrid Mamba-2 + Attention) MoE language backbone with a RADIO v2.5-H vision encoder and a Parakeet (FastConformer) sound encoder, supporting interleaved text, image, and audio inputs.

Task

Omnimodal (Text·Image·Audio)

Architecture

NemotronH_Nano_Omni_Reasoning_V3

Parameters

30B total / 3B active

HF Org

nvidia

Available Models#

  • Nemotron-3-Nano-Omni-30B-A3B-Reasoning: 30B total, 3B activated (MoE)

Architecture#

  • NemotronH_Nano_Omni_Reasoning_V3

Example Recipes#

Recipe

Dataset

Description

nemotron_omni_cord_v2.yaml

CORD-v2

Full SFT — receipt parsing

nemotron_omni_cord_v2_peft.yaml

CORD-v2

LoRA PEFT — receipt parsing

Try with NeMo AutoModel#

1. Install (NeMo AutoModel):

pip install nemo-automodel

2. Clone the repo to get the example recipes:

git clone https://github.com/NVIDIA-NeMo/Automodel.git
cd Automodel

3. Run the recipe from inside the repo (8x H100 example):

automodel examples/vlm_finetune/nemotron_omni/nemotron_omni_cord_v2.yaml --nproc-per-node 8

For a full walkthrough — dataset preparation, SFT vs. LoRA configs, and post-training inference — see the Nemotron-Omni guide.