#6 Open Source Forum 2019 Adam Gibson
summary
Overview of concrete engineering issues: claims that framework choice has consolidated to a TensorFlow/PyTorch duopoly but both are C/C++ codebases under Python, that deploying models needs nontrivial graph massaging and vendor-specific hacks, and that Kubernetes and mainstream schedulers poorly support custom accelerators. Tools and standards mentioned: ONNX, TVM/MLIR, HDF5, FlatBuffers; packaging work includes a Java/C++ bridge for TensorFlow used since 2012 and distributed training over Apache Spark with Aeron. Watch if you want practical notes on portability, serving, and metricization (Prometheus) rather than high-level theory.
AI-written summary · about
chapters (6)
related talks
05mThe future of frameworks: Navigate the OSS landscape
37mPanel Discussion
11mOpen Source for the AI Era
16mKeynote: Open Source for AI and AI for Open Source
45mKeynote: Welcome
45mValues-based OSS Open Discussion
17mOpen Innovation: Unlocking AI's Potential
12mCo-Evolution: How the Open Source Intelligence Stack Compounds