Serverless Tensorflow 1.0.0 on AWS. Wait what?
summary
Ports TensorFlow 1.0 / 1.4 and Keras models into AWS Lambda using Docker on Amazon Linux, building a deployment artifact in the lambda-packs repo and loading models from S3. Demonstrates Inception v3 image captioning that runs in about four seconds, costs $0.00005 per run (≈20,000 images per $1) and scales to thousands of concurrent functions. Notes limits: ~500 MB package, ~1.5 GB RAM, 5 min timeout, CPU-only; solution works but needs model size and startup optimization. lambda-packs on GitHub
AI-written summary · about
chapters (7)
- 0:00Introduction to serverless
- 2:24How AWS Lambda works
- 4:12Why TensorFlow on Lambda
- 5:40Porting TensorFlow limitations
- 7:09Real-world implementations
- 8:13Demo and code walkthrough
- 11:05Q&A and closing remarks
related talks
12mServerless Deep Learning
18mServerless AI's Rise: Revolutionizing ML Deployment
46mServing machine learning models as an inference API in production
23mOn-demand Container Loading in AWS Lambda
54mContainerless in the cloud with AWS Lambda
34mBuilding scalable end-to-end deep learning pipelines in the cloud
32mServerless Data Processing and Machine Learning
41mBuilding Serverless Machine Learning Models in the Cloud