Browse Skills

cuOpt Routing Python API solves vehicle routing problems (TSP, VRP, PDP) with NVIDIA cuOpt from Python. Use it when building or solving routing models in Python.

cuOpt Server API Python provides REST server deployment and Python/curl client examples for NVIDIA cuOpt routing and LP/MILP problems. Use it when starting the server, understanding endpoints, or writing client code.

cuPyNumeric HDF5 I/O reads and writes large cuPyNumeric arrays to HDF5 files using Legate's parallel, distributed HDF5 I/O. Use it when a developer needs to save or load .h5/.hdf5 datasets, read large files in chunks, or accelerate HDF5 disk I/O with GPUDirect Storage.

cuPyNumeric Install helps set up and verify the cuPyNumeric Python library via conda or pip, including GPU usage checks. Use it when someone needs cuPyNumeric running in an isolated environment on Linux or WSL.

cuPyNumeric Migration Readiness assesses NumPy code before porting to cuPyNumeric, identifying which APIs will scale on GPU and which need refactoring. Use it before starting a migration to get a verdict and concrete refactor pointers.

CuPyNumeric Parallel Data Load loads sharded, on-disk datasets (.npy, Parquet/Arrow, raw binary, HDF5) into a distributed cuPyNumeric ndarray using manual partition and leaf task launch for CPU, OMP, and GPU variants. Use it when no single-call loader fits, including when per-shard row counts differ across files.

DALI Dynamic Mode guides AI agents in writing, reviewing, and migrating code that uses NVIDIA DALI's imperative dynamic-mode API (`nvidia.dali.experimental.dynamic`, ndd). Use it when working on ndd code or migrating pipeline-based DALI code to dynamic mode.

Data Designer builds synthetic datasets and data generation pipelines using the Data Designer library. Use it when you need to generate structured data, create datasets, or build a data generation pipeline.

DeepStream Development builds NVIDIA DeepStream SDK pipelines using the Python pyservicemaker API for video analytics, including object detection, tracking, and TensorRT inference. Use when creating GStreamer-based video processing, integrating inference models, or adding Kafka/message broker support.

DeepStream Import Vision Model automates the end-to-end import of object detection vision models from HuggingFace or NVIDIA NGC into an NVIDIA DeepStream pipeline, including ONNX download or SafeTensors export, TensorRT engine build, custom parser, multi-stream benchmark, and PDF report.

DeepStream Pipeline Builder generates ready-to-run gst-launch-1.0 pipelines for NVIDIA DeepStream SDK. Use when the user asks about pipelines for video/image inference, detection, tracking, or streaming.

DeepStream Profiling profiles NVIDIA DeepStream pipelines using Nsight Systems and derives performance configs such as batch size and precision from measurements. Use it when building efficient pipelines or benchmarking FPS on a target GPU.