FAQ for DeepStream GitHub Mono-repo#
1. Where is the DeepStream GitHub mono-repo located?#
Starting with DeepStream 9.0, this mono-repo is the unified home for DeepStream source, samples, tools, and skills across x86, Jetson, and SBSA. SDK release packages moved to the mono-repo GitHub release assets in DeepStream 9.1 (see Q3).
2. Where can I download the DeepStream SDK packages?#
Starting from DeepStream 9.1, all DeepStream SDK packages are available as release assets on the NVIDIA DeepStream GitHub mono-repo:
Navigate to the release corresponding to your DeepStream version and download the required package from the Assets section.
3. Why has DeepStream moved from NGC to GitHub for release packages?#
The NVIDIA/DeepStream GitHub mono-repo is the unified platform for all DeepStream releases. DeepStream 9.0 introduced the mono-repo as the release channel, but SDK packages (assets) were still hosted on NGC. Starting from DeepStream 9.1, the mono-repo GitHub releases page is the single place for both the release and all downloadable SDK packages. NGC is no longer used for SDK package distribution.
4. What packages are available in the GitHub release assets?#
The following packages are available for each DeepStream release:
Full DeepStream package (tar and deb) — complete SDK for x86 and aarch64
deepstream-binaries-x86 (tar and deb) — proprietary binary libraries only, for x86
deepstream-binaries-aarch64 (tar and deb) — proprietary binary libraries only, for aarch64
deepstream-samples (tar and deb) — sample streams, configs, and models
DeepStream Libraries wheel (
.whl) — the DeepStream Libraries Python wheel (e.g.deepstream_libraries-1.4-cp312-cp312-linux_x86_64.whl), for x86
5. What is the difference between the full DeepStream package and the deepstream-binaries package?#
Full DeepStream package: Contains the complete SDK — all open-source components and proprietary libraries — for users who want to install DeepStream directly.
deepstream-binaries package: Contains only the proprietary binary libraries. This is intended for users who build DeepStream from source using the mono-repo and need only the closed-source components to complement their build.
6. Where can I find the NVIDIA-AI-IOT GitHub repositories in the GitHub mono-repo?#
The table below maps the existing NVIDIA-AI-IOT GitHub repository sources to their location within the GitHub mono-repo:
NVIDIA-AI-IOT GitHub repo |
Location within the GitHub mono-repo |
|---|---|
7. Where do I get the DeepStream Libraries Python wheel?#
Starting with DeepStream 9.1, the DeepStream Libraries Python wheel is published as a release asset on the mono-repo GitHub releases page, alongside the other SDK packages:
Download the wheel (for example deepstream_libraries-1.4-cp312-cp312-linux_x86_64.whl) from the Assets section and install it with pip3 install <wheel>. It was previously distributed on NGC; NGC is no longer used for the DeepStream Libraries wheel. The DeepStream Libraries source and sample applications are hosted as a submodule of the mono-repo at NVIDIA/DeepStream. See DeepStream Libraries for full setup and usage details.
8. How do I download DeepStream release assets from GitHub using wget or curl?#
You can download release assets directly from the command line using curl or wget.
Using curl:
$ curl -LO 'https://github.com/NVIDIA/DeepStream/releases/download/v9.1.0/deepstream-9.1_9.1.0-1_amd64.deb'
Using wget:
$ wget --content-disposition 'https://github.com/NVIDIA/DeepStream/releases/download/v9.1.0/deepstream-9.1_9.1.0-1_amd64.deb'
Replace the filename with the specific asset you want to download from the release page.
9. Can I still download DeepStream from NGC?#
Starting from DeepStream 9.1, NGC is no longer the distribution channel for DeepStream SDK packages. All packages are now available through the GitHub releases page at NVIDIA/DeepStream.
However, DeepStream Docker containers continue to be available on NGC at https://catalog.ngc.nvidia.com/orgs/nvidia/containers/deepstream.
10. Where did the DS 9.1 public tar/deb files go since they are no longer on NGC?#
They are now located in the DeepStream mono-repo. See the DS 9.1 release tag:
11. The Jetson DS 9.1 containers are not found in the NGC deepstream-l4t repo (https://catalog.ngc.nvidia.com/orgs/nvidia/-/containers/deepstream-l4t). Where did they go?#
The deepstream-l4t container NGC repo will not be updated in future DeepStream releases, though old releases will remain accessible to users.
Starting with the DS 9.1 release, the DeepStream container NGC repo https://catalog.ngc.nvidia.com/orgs/nvidia/-/containers/deepstream is the official repo for all containers (Jetson and dGPU).
12. Where did the DeepStream Open Source Dockers move to for DS 9.1? (previously NVIDIA-AI-IOT/deepstream_dockers)#
13. Where are DeepStream Skills, auto-magic-calib, and inference-builder located with GitHub Monorepo?#
These repos are now also hosted into the mono-repo as listed below:
DeepStream Skills: NVIDIA/DeepStream
auto-magic-calib (Original repo NVIDIA-AI-IOT/auto-magic-calib): NVIDIA/DeepStream
inference-builder (Original repo NVIDIA-AI-IOT/inference_builder): NVIDIA/DeepStream
14. Are there any proprietary files in the DeepStream mono-repo?#
Yes, please refer to the EULA file packaged within the proprietary tar/deb released as GitHub Release assets.
Note
The open-source components in the mono-repo are licensed under the Apache License 2.0.
15. I am observing a warning while launching dockers: “groups: cannot find name for group ID 993” or similar. Where does this come from?#
This is just a warning. It was introduced in nvidia-container-toolkit 1.19.1-1. For more details, see:
This applies to all platforms (x86/Jetson/SBSA).
16. How do I build or modify DeepStream components on SBSA / DGX Spark?#
DeepStream SDK release assets are not published for SBSA/DGX Spark. Use the official NGC Docker container, which ships with DeepStream fully installed and all dependencies pre-configured:
To modify a specific open-source component, clone the mono-repo inside the container and run make only in that component’s directory. Since all prerequisites are already present in the container, there is no need to run the full build/build.sh:
$ git clone https://github.com/NVIDIA/DeepStream.git
$ cd DeepStream/src/<component-directory>
$ make && sudo make install