Polyglot Tensorflow: integrating Go and Python
summary
NeuralMed migrated TensorFlow inference from Go to Python because the Go TensorFlow SDK lacked maintenance; they validated models in Python but then faced integration choices (synchronous vs asynchronous communication) and transport options (HTTP/REST, gRPC, or messaging). The talk explains how they evolved a polyglot architecture combining Go and Python components to provide stable inference serving for multiple deep-learning models.
AI-written summary · about
chapters (6)
- 0:00Start
- 21:43Validation of results
- 32:38Evolution of multiple deep learning models
- 36:04Activation maps
- 37:15Criticality levels
- 40:45Key points
