Find starting configurations for training MoE models, which route work among specialized expert networks, on H100, B200, GB200, and GB300. Provides model-specific parallel layouts, communication options, and approximate performance bands to guide measured comparisons.
Write and review NeMo-RL documentation using its project conventions. Keep the documentation index aligned with added or renamed Markdown files, document new features in the appropriate location, and format function and class docstrings in Google style.
Turn a Jetson video encoding request into a structured JSON recipe with configurations for Video Codec SDK and PyNvVideoCodec. Record defaults, explicit requirements, and settings that either interface cannot represent, without running an encode.
Understand the Isaac for Healthcare workflow architecture and discover supported workflows and run modes from the current checkout. Get directed to the appropriate stage skill for setup, data processing, training, or validation.
Collects a read-only health snapshot of a Jetson device, covering its model, memory, GPU, temperatures, power mode, storage, services, and processes. Helps investigate slowdowns, overheating, and memory pressure using observed device data.
Write a prepared board support package image to a Jetson device in recovery mode. The skill checks device identity and image readiness, selects the flashing tool for the boot storage, and presents the resolved plan for approval.
Choose containers and Python package sources that match a Jetson device's GPU architecture, JetPack, and CUDA environment. Get compatibility hints for PyTorch, vLLM, and related stacks before installation, using NVIDIA catalogs and Jetson AI Lab indexes.
Install, repair, or inspect NVIDIA Video Codec SDK and PyNvVideoCodec on Jetson. Verify readiness with official samples that encode and independently decode one frame, and report what passed, what failed, and the next action.
Investigate a failed GitHub Actions job and turn its logs into a structured bug issue for Megatron-LM. Collects failing tests, errors, and related pull requests, checks for duplicates, and identifies an assignee from test history.
Switch the active Jetson target to an existing YAML profile before customization, building, or flashing. Review profile names and configuration summaries, choose a target, and update the pointer used by downstream skills without creating new profiles.
Collect a short, read-only report from a running Jetson device, covering its model, L4T release, kernel, OS, power mode, and uptime. Use it to record the environment before performance, regression, or compatibility tests.
Launch, monitor, stop, and troubleshoot NeMo-RL training jobs on Kubernetes using nrl-k8s. Choose a temporary cluster for a single run or reuse a persistent cluster while adjusting training settings and investigating failures.
Start Jetson or IGX board software customization by selecting a setup mode, platform, release, and carrier-board intent in a questionnaire. It passes those choices to the appropriate setup skills and reports workspace readiness and remaining steps.
Prepare configuration files for a custom Jetson carrier board using a reference development kit. Creates board-specific copies and a device tree overlay template, a file for hardware configuration differences, with build registration and flash configuration wiring.
Check and format Python code in Megatron-LM using autoformat.sh and its linting tools. Apply project rules for imports, type hints, documentation, naming, and line length when preparing changes for a pull request.
Measure video encoding or decoding throughput on Jetson, compare presets, and test concurrent worker capacity. When representative media is unavailable, produce a clearly labeled estimate from SDK documentation, keeping estimates separate from measured results.
Choose software for serving language or vision-language models on NVIDIA Jetson using a current memory audit. Get specific launch flags for memory limits, context length, and concurrency to address memory shortages or compare lower-memory alternatives.
Configure how Jetson Orin or Thor assigns shared high-speed connections on a custom carrier board. The skill checks board schematics and platform guides, updates ODMDATA settings, and coordinates controller-specific device tree changes through related skills.
Run Isaac for Healthcare workflows in a simulator using a learned policy or rule-based controller. Verify completed episodes through the final success summary, visible behavior, and recorded data, with checkpoint compatibility checks when a checkpoint is supplied.
Check Jetson video encoding and decoding capabilities by comparing live API reports, verified NVIDIA samples, and official documentation. Distinguish documented product support from observed behavior for specific formats, bit depths, dimensions, and tested operations.
Reclaim memory on Jetson devices that do not need a local desktop or display. Build a plan from current memory audit data, preview reversible service changes, apply approved changes, and measure the actual memory savings afterward.
Guide checks of a customized Jetson system image, from file consistency to tests on a flashed device over SSH or UART. Device access is defined, while the broader validation procedure remains a skeleton.
Rebuild Jetson hardware configuration files, kernel images, and modules after source changes. Selects a build mode from the changed repositories or an explicit request, then records the outputs in a manifest for a separate deployment step.
Manage nvfancontrol fan profiles for Jetson and Tegra boards. Add, remove, or edit temperature-based fan curves and boot defaults in board support files, while tracking changes and preserving configuration differences between related carrier boards.
Is this your account? Link GitHub to bring it onto your creator page (Japanese)