tao-train-nvdinov2
nvidiaBuild visual features without labeled data.
visual representation learning
What it does
Trains, exports, and runs inference for a TAO NVDINOv2 backbone using self-supervised teacher-student learning.
When to use it
Use when training, exporting, or running inference for a TAO NVDINOv2 backbone.
How to use it
Give it a TAO NVDINOv2 train, export, or inference request with its data and model inputs; it runs the corresponding configured action and produces the specified results.
What you provide
- the requested TAO action
- dataset archives
- checkpoint or pretrained weights
- An existing project
- Your files
Access · 0
No additional actions listed in the analysis.
What you need · 4
Docker must be available for the skill to run.
NVIDIA Container Toolkit must be available for the skill to run.
The skill requires A100 GPUs with at least 40GB of VRAM per GPU, with a minimum of four GPUs and eight recommended.
Training defaults to tao-run-automl when AutoML is enabled and the packaged train schema and template are present; direct training is used when AutoML is off or unavailable.
About this skill
- Visibility
- Public
- Repository
- nvidia/skills
- Created
- Oct 8, 2026
- Updated
- Oct 8, 2026
- Files
- 18