Foundation models come with broad language and reasoning capabilities across use cases and modalities based on the training datasets used. Models understand language and can follow instructions, but specialized workflows often require context that is restricted, specialized, or proprietary. Customizing an agent solves this challenge by shaping how the agent reasons under constraints, which tools it selects, how it structures its outputs, and how reliably it executes domain workflows.
Mastering Agentic Techniques: AI Agent Customization | NVIDIA Technical BlogIn practice, the most effective agent customization combines multiple techniques in sequence. The stages of a representative pipeline are outlined below. Start with system prompts, tool and skill definitions, and retrieval to establish baseline behavior.
Mastering Agentic Techniques: AI Agent Customization | NVIDIA Technical Blog… This photorealistic data is crucial for post-training policy models, ensuring smooth simulation-to-reality transfer and supporting model training for perception AI and specialized robot models like GR00T N1 . …
… We show the local training code on the left and the federated version on the right, highlighting: import, flare.init , receive , send . train.py train.py import torch import torchvision import torchvision.transforms as transforms from model import Net batch size = 4 epochs = 1 lr = 0.01 model = Net… …
… When to use Working with accessible data for well-defined tasks with output examples Customizing a model for a low-resource domain where labelled examples are limited, and high-quality synthetic data can be generated to bootstrap the fine-tuning dataset Requiring the model to reliably produce speci… …
… The SDG pipelines are also available as data designer recipes . The image training data is permissively released at huggingface.co/datasets/nvidia/Nemotron-Image-Training-v3 . With the underlying image data and the model, developers can inspect, adapt, and extend multimodal training pipelines. …
… 3a Replay demos sanity check ./workflows/rheo/docker/run docker.sh -g1.6 \ python scripts/sim/replay demos.py \ --dataset file /datasets/demo.hdf5 \ --enable cameras \ g1 locomanip tray pick and place \ --object surgical tray \ --embodiment g1 wbc pink 3b Annotate demos prepare for generation ./wor… …
… Training data for stronger agent reasoning As with all Nemotron open model launches, much of the training data pipeline is released as permissively as possible. For partners in enterprise and sovereign AI development, training data transparency and provenance matter as much as capability. …
… Solve the Training Data Challenge With the Physical AI Data Factory Learn how to build a training data pipeline with your own video or image data. Then, curate, augment, and evaluate it with Cosmos for vision AI models. …
… This post compares the following three low-precision training formats directly against established BF16 precision training across multi-hundred-billion token pretraining runs and downstream benchmarks: 8-bit floating point per-tensor current scaling FP8-CS Mixed precision training with FP8 MXFP8 NV… …
… This helps accelerate DLI applications by reducing data access latency and training time. Step 3: Simulation and Testing Simulation Simulation of real-world scenarios isn't only critical for training for unforeseen scenarios, but also vital for testing and validation of autonomous vehicles. …
… Nemotron Omni Datasets Multimodal data extending the Nemotron training pipeline beyond text to image, video, and speech. ~127B tokens of cross-modal pretraining data and ~124M curated post-training examples for document reasoning, computer use, and long-horizon workflows. …