Accelerating ML Prototyping The Pythonic Way
summary
Meta runs an internal Jupyter distro Bento that adds scheduled notebooks for recurring reports, persistent sessions and notebook recovery so long-running executions resume after disconnects, and one‑click cross-platform debugging between VS Code and notebooks. A portable “k” virtual environment surfaces GitHub workflows inside the monorepo while conda handles dependency isolation, telemetry, error and vulnerability tracking. The team reports these changes cut prototyping friction and is experimenting with using k as a notebook/kernel bridge to push research environments closer to production.
AI-written summary · about
related talks
21mMeta: Enabling Programming Languages to Learn from the Crowd
30mDrops of Jupyter: An Exploration of the Notebook Ecosystem
24mBuilding an ML Application Platform from the Ground Up
27mNotebook To Production
22mImprove Your ML Projects: Embrace Reproducibility and Production...
14mReal-Time Data Processing for ML Feature Engineering
30mHuman-Centric ML Infrastructure at Netflix
35mBuilding Robust Reactive ML Systems In A Multiplayer Setting