Building secure and minimalistic Docker images for Data and ML with Rockcraft
summary
Workshop shows how to build OCI “rocks” with Rockcraft, slice Debian packages with Chisel and load results into Docker with Skopeo; exercises produce a working numpy image that imports correctly and a PyTorch+Jupyter rock that runs as a Pebble service. Build sequence: full Ubuntu base → 61 MB compressed (172 MB uncompressed), switch to be-base reduces to 37 MB (101 MB uncompressed), then chisel cut slicing shrinks to 30 MB (82.8 MB uncompressed). Commands used include rockcraft pack, chisel cut, and skopeo to export.
AI-written summary · about
chapters (9)
- 0:00Workshop introduction
- 1:05Setting up requirements and workshop files
- 5:11Rocks and Rockcraft
- 6:54Chisel and package slicing
- 8:44Hands-on NumPy Rock
- 27:20Minimal images with bare base
- 31:38Chisel slices optimization
- 54:48PyTorch and Jupyter Rock
- 1:19:55Q&A
related talks
22mPush straight to prod: API development with R and Tensorflow
23mBringing Container-Native Simplicity to AI/ML
54mRipped Machine Learning with Distributed Computing
24mDocker for Machine Learning!
18mBaby Sitting Your AI Models in Production? STOP!
33mDeploying and Scaling Spark ML and Tensorflow AI Models
01h 30mUnleashing the Power of AWS: A Workshop on Bedrock Assistant
16mReducing Operational Barriers to Model Training