Devcontainers for AI/ML Workflow Optimization
summary
Claims reproducible AI/ML environments across AMD and NVIDIA hardware by using dev containers and an open dev container template checked into Git. Demo uses ROCm 5.2 and PyTorch 2.9.1, shows a container recognizing the GPU and running hello_gpu.py, and shows uv only + pyproject.toml protecting preinstalled packages so running uv add transformers does not overwrite the accelerator PyTorch. Workflow goes from local dev container to a production Dockerfile and Kubernetes image.
AI-written summary · about
chapters (9)
- 0:00AI/ML dependency challenges
- 3:47What are containers
- 6:56Dev containers simplify dev
- 13:12Live demo setup AMD ROCm
- 17:23Creating dev container
- 19:27Testing GPU acceleration
- 20:34Managing dependencies with UV
- 21:38Path to production
- 23:04Summary next steps
related talks
24mDocker for Machine Learning!
28mContainers: the Last Opportunity to Make Reproducible AI
32mDeveloping on Containers with Dev Containers
01h 02mDocker for Data Science: Reproducibility and Deployment
22mDock Your Jupyter Notebook
19mDepend on Docker for Self-Managed Predictive Modeling
28mEternal sunshine of the spotless development environment
56mDevOps for Machine Learning