PyTorch on Google Cloud: Blog series recap

PyTorch is an open supply machine studying framework, primarily developed by Meta (beforehand Fb). PyTorch is  extensively  used within the analysis area  and in recent times it has gained immense traction within the business resulting from its ease of use and deployment. Vertex AI, a totally managed end-to-end knowledge science and machine studying platform on Google Cloud, has top quality assist for PyTorch making it optimized, compatibility examined and able to deploy. 

We began a brand new weblog sequence – PyTorch on Google Cloud – to uncover, show and share how one can construct, practice and deploy PyTorch fashions at scale on Cloud AI Infrastructure  utilizing GPUs and TPUs on Vertex AI, and how one can create reproducible machine studying pipelines on Google Cloud . This weblog put up is the house web page to the sequence with hyperlinks to the present and upcoming posts for the readers to seek advice from. Listed below are hyperlinks to the weblog posts on this sequence:

PyTorch on Vertex AI

1. How To coach and tune PyTorch fashions on Vertex AI: On this put up, discover ways to use Vertex AI Coaching to construct and practice a sentiment textual content classification mannequin utilizing PyTorch and Vertex AI Hyperparameter Tuning to tune hyperparameters of PyTorch fashions

2. How one can deploy PyTorch fashions on Vertex AI: This put up walks by means of the deployment of a Pytorch mannequin utilizing TorchServe as a customized container by deploying the mannequin artifacts to a Vertex Prediction service.

3. Orchestrating PyTorch ML Workflows on Vertex AI Pipelines: On this put up, we present how one can construct and orchestrate ML pipelines for coaching and deploying PyTorch fashions on Google Cloud Vertex AI utilizing Vertex AI Pipelines.

4. Scalable ML Workflows utilizing PyTorch on Kubeflow Pipelines and Vertex Pipelines:  This put up exhibits examples of PyTorch-based ML workflows on two pipelines frameworks: OSS Kubeflow Pipelines, a part of the Kubeflow challenge; and Vertex AI Pipelines. We share new PyTorch built-in parts added to the Kubeflow Pipelines. 

PyTorch/XLA and Cloud TPU

1. Scaling deep studying workloads with PyTorch / XLA and Cloud TPU VM: This put up describes the challenges related to scaling deep studying jobs to distributed coaching settings, utilizing the Cloud TPU VM and exhibits how one can stream coaching knowledge from Google Cloud Storage (GCS) to PyTorch / XLA fashions operating on Cloud TPU Pod slices

2. PyTorch/XLA: Efficiency debugging on Cloud TPU VM: Half I: On this first a part of the efficiency debugging sequence on Cloud TPU, we lay out the conceptual framework for PyTorch/XLA within the context of coaching efficiency. We launched a case research to make sense of preliminary profiler logs and establish the corrective actions.

3. PyTorch/XLA: Efficiency debugging on Cloud TPU VM: Half II: Within the second half,  we deep dive into additional evaluation of the efficiency debugging to find extra efficiency enchancment alternatives.

4. PyTorch/XLA: Efficiency debugging on Cloud TPU VM: Half III: Within the remaining a part of the efficiency debugging sequence, we introduce consumer outlined code annotation and visualize these annotations within the type of a hint.

5. Prepare ML fashions with Pytorch Lightning on TPUs: This put up exhibits how straightforward it’s to begin coaching fashions with PyTorch Lightning on TPUs with its built-in TPU assist.

Just a few extra articles associated to PyTorch on Google Cloud

1. How To coach PyTorch fashions on AI Platform: On this put up, discover ways to setup PyTorch growth surroundings on Vertex AI Workbench (beforehand Notebooks) and use AI Platform Coaching to construct and practice a sentiment textual content classification mannequin utilizing PyTorch.

2. Enhance your productiveness utilizing PyTorch Lightning: Learn to use PyTorch Lightning on Vertex AI Workbench (beforehand Notebooks)

We’ve got extra articles coming quickly protecting PyTorch and Google Cloud AI. 

Keep tuned. Thanks for studying! Have a query or need to chat? Discover us right here: Vaibhav Singh, Rajesh Thallam, Jordan Totten and Karl Weinmeister.