Training AI To Code Using The Largest Code Dataset (Project CodeNet)
summary
Project CodeNet is a 14 million–sample, 500 million–line corpus across 55 languages that the team uses with Project CodeNet subsets, DataShim and the Machine Learning Exchange to train and serve models on Kubernetes. They show a masked language model trained on a C subset (50k examples, 5 epochs) achieving ~92% top‑1 and ~99% top‑5 on masked-token prediction. Pipelines are auto-generated as Kubeflow-on-Tekton, models are containerized and served (KServe), and the end-to-end flow demonstrably runs on cluster infrastructure.
AI-written summary · about
chapters (5)
- 0:00Start
- 13:50Mass Language Models
- 19:04Machine Learning Exchange
- 20:21Api Layers
- 28:08Language Classification Models
related talks
37mTraining AI To Code Using the Largest Code Dataset
39mTraining AI To Code Using The Largest Code Dataset (Project CodeNet)
37mMachine Learning on Source Code
28mDelivering AI on Code: Live Demo of source{d}
46mCode + AI: Will Robots Take Our Coding Jobs? ML Applied to Programming
45mML: ML.NET: deep dive into dev AI
58mExploring Machine Learning for .Net Developers
40mEnd-to-end AI Marketplace with Execution Engine on Kubernetes