Best Practices for Building Production-Grade Deep Learning Systems
summary
Demonstration shows concrete paths to production: use BERT on MXNet Gluon trained on movie reviews to 94–95% test accuracy, running on p3.8xlarge GPUs in about 20–30 minutes; compare using preconfigured Deep Learning AMIs versus SageMaker training + REST endpoint (SageMaker provisions and tears down clusters); pipeline tools highlighted include Ground Truth, Deep Learning Containers, Elastic Inference and container orchestration for MLOps.
AI-written summary · about
chapters (19)
- 0:00Introduction
- 0:41Deep Learning
- 1:18Deep Learning Production
- 2:50Deep Learning Workflow
- 3:59Data Science Workflow
- 6:24ML Ops
- 7:00Production Deep Learning
- 9:28AWS Deep Learning Ecosystem
- 11:06Multilingual Deep Learning Developers
- 11:21Building on AWS
- 12:13SageMaker
- 13:19AWS Console
- 14:52NLP Architecture
- 15:16SSH
- 15:59Running the Model
- 18:53Jupiter Lab
- 19:18Importing Libraries
- 20:05Setting Up Data
- 23:01Outro
related talks
40mHow to Implement a Production Ready Deep Learning Model in 12 Weeks
43mDesigning and Building Complex Machine Learning Projects
01h 16mDeep Dive into Deep Learning with MXNet
54mDeep learning on AWS with NVIDIA: From training to deployment
35mAccelerate deep learning in the cloud with custom ML environments
44mScaling ML from 0 to millions of users
45mBuild NLP Models with Amazon SageMaker
44mDeep Learning in Production