Official NVIDIA-authored guidance for NVIDIA cuDF GPU DataFrames, pandas acceleration, dask-cuDF, ETL, joins, groupby, CSV/Parquet I/O, nullable semantics, and multi-GPU DataFrame workloads.
日本語の概要は準備中です。原文の説明を表示しています。
Use when you need to print Jetson device info (module model, L4T version, kernel, OS version, current power mode) from a running Jetson target. This is an example skill.
インストールする前に、エージェントに与えられる指示の中身を確認できます。
Prints a concise summary of the Jetson device this skill runs on.
This skill is intended as a reference example for the jetson-device-skills repo and the NVIDIA-wide skills CI. It runs on the Jetson target (not the host PC) and performs read-only inspection — useful as a baseline capture before running performance tests.
Capture a baseline snapshot of a Jetson target's software stack and power mode before running performance, regression, or compatibility tests.
tr, cat, uname, uptime, lsb_release (or /etc/os-release).nvpmodel available for power mode (optional — skill falls back gracefully).None. The skill reads only from the local Jetson filesystem and standard CLIs.
Run each step in order and print the captured values into the report shown under Output format.
# Device-tree strings are null-terminated, so strip NULs before printing.
model=$(tr -d '\0' < /proc/device-tree/model 2>/dev/null)
case "$model" in
*Jetson*) ;;
*) echo "Not running on a Jetson target (model: '${model:-unknown}')"; exit 1 ;;
esac
echo "$model"
head -1 /etc/nv_tegra_release 2>/dev/null || echo "L4T release info not found"
# Equivalent: grep -m1 '^# R' /etc/nv_tegra_release
nvpmodel -q and join its two output lines (NV Power Mode: <name> and the mode number) onto one line. paste -sd only uses the first char of its delimiter, so use awk to insert the literal / separator:
nvpmodel -q 2>/dev/null | awk 'NR==1{a=$0; next} {print a" / "$0}' \
|| echo "nvpmodel not available"
uname -r
uptime -p
lsb_release, fall back to /etc/os-release):
lsb_release -ds 2>/dev/null || (. /etc/os-release && echo "$PRETTY_NAME") || echo "OS version not found"
Print a short report with these sections, one line each where possible:
Model: <device-tree model string>
L4T release: <release header line>
Power mode: <nvpmodel name> / <mode number>
Kernel: <uname -r>
Uptime: <uptime -p>
OS version: <lsb_release / /etc/os-release output>
Example output on an Orin AGX dev kit (L4T R36 / Ubuntu 22.04):
Model: NVIDIA Jetson AGX Orin Developer Kit
L4T release: # R36 (release), REVISION: 3.0
Power mode: NV Power Mode: MAXN / 0
Kernel: 5.15.136-tegra
Uptime: up 2 hours, 14 minutes
OS version: Ubuntu 22.04.4 LTS
Example output on an AGX Thor (L4T R39 / Ubuntu 24.04):
Model: NVIDIA Jetson AGX Thor Developer Kit
L4T release: # R39 (release), REVISION: 0.0
Power mode: NV Power Mode: 120W / 1
Kernel: 6.8.0-tegra
Uptime: up 12 minutes
OS version: Ubuntu 24.04 LTS
Each command falls back to a clearly labeled "... not found" / "... not available" string if the underlying file or binary is missing — the skill never errors out mid-report. If /proc/device-tree/model is missing or does not contain a Jetson string, exit early with a clear "not running on a Jetson target" message.
nvpmodel output format varies between L4T versions; the skill prints it verbatim rather than parsing.Error: /proc/device-tree/model: No such file or directory
Cause: Running on a host PC, not a Jetson target.
Solution: Run the skill on the Jetson device directly (e.g. via SSH).
Error: nvpmodel: command not found
Cause: L4T BSP not installed, or running in a minimal container without nvpmodel.
Solution: Expected on non-Jetson or stripped environments — the skill prints "nvpmodel not available" and continues.
/proc/device-tree/model will not contain a Jetson model string — detect that and exit with a clear message rather than printing misleading info.まだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
Official NVIDIA-authored guidance for NVIDIA cuDF GPU DataFrames, pandas acceleration, dask-cuDF, ETL, joins, groupby, CSV/Parquet I/O, nullable semantics, and multi-GPU DataFrame workloads.
日本語の概要は準備中です。原文の説明を表示しています。
Customize NVIDIA Nemotron Voice Agent's Generic Pipecat example for healthcare appointment, five-field patient intake, or custom tool-calling workflows without a separate backend.
日本語の概要は準備中です。原文の説明を表示しています。
Calibrate a new dataset from live RTSP camera streams via the AutoMagicCalib REST API. Use when the user provides RTSP URLs or asks to calibrate live cameras; VIOS records clips, AMC ingests them, then runs calibration.
日本語の概要は準備中です。原文の説明を表示しています。
Run end-to-end calibration on the shipped sample dataset (sdg_08_2_sample_data_010926.zip) against a running AMC microservice. Use when user says 'test sample dataset', 'run sample calibration', 'verify AMC install', or 'launch and test'.
日本語の概要は準備中です。原文の説明を表示しています。
Calibrates pre-recorded `cam_*.mp4` datasets through the AutoMagicCalib REST API. Use for user-supplied local MP4s; route live RTSP streams to `amc-run-rtsp-calibration`.
日本語の概要は準備中です。原文の説明を表示しています。
Launch AutoMagicCalib microservice and web UI from NGC release images via Docker Compose. Use when user says 'deploy auto calibration', 'launch auto calibration', 'launch AMC', 'start MS+UI', or 'set up auto-magic-calib'. Requires NGC API key.
日本語の概要は準備中です。原文の説明を表示しています。