Bring the Power of ONNX to Spark as it Never Happened Before
summary
Adds an ONNX Runtime execution provider for Huawei Ascend so ONNX models run directly on Ascend hardware via CANN/ATC/TBE instead of model translation, with CI pushed upstream. Demonstrates ResNet and VGG run correctly on Atlas 300 PCI cards, and a roadmap to support all ONNX operators by year end plus follow‑up optimizations. Shows a Spark integration using a model UDF and the Spark SPIP simplified API to execute ONNX inference at scale over DataFrame pipelines.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 1:04Spark and deep learning
- 6:08Spark AI improvements
- 13:50ONNX basics
- 17:45Huawei Ascend ecosystem
- 23:26ONNX runtime on Ascend
- 26:01Spark ONNX inference
- 28:13Q&A
related talks
30mScalable ML acceleration with ONNX Runtime
40mScala and Spark in the New Deep Learning Era
30mThe Evolution of ONNX: Driving Interoperability in AI...
45mLeverage Power of Machine Learning with ONNX
36mApplied AI on the edge
42mAI Everywhere– accelerate your development from PC to cloud
31mScoring ONNX ML Models with Scala
05mONNX Runtime