Developers can adapt the model to domain workflows using NeMo training tools and optimized inference recipes. Core capabilities include Text and image understanding, Document comprehension, Video understanding, and Reasoning and instruction following.
Public code or model weights are linked from the project, with setup instructions for running or adapting the system. Additional characteristics include Tool use, NeMo fine-tuning, NVFP4 optimization recipes, and Context reuse and speculative decoding.
