ml-base

Docker images for running PyTorch and Cuda-based workloads. These can be run locally or in the cloud on something like GKE.

• Dockerfile.base - Inherits from gcr.io/deeplearning-platform-release/pytorch-gpu and installs further dependencies
• Dockerfile.runtime - Intended as a lighter weight inference-only image. To be used for APIs, etc.
• Dockerfile.training-base - Use for training new embeddings or models based on the huggingface transformer architecture.