Systems | Development | Analytics | API | Testing

July 2021

[MLOps] The Clear SHOW - S02E13 - mlops_this: Copilot Shenanigans

Ariel should have known better than to mess with shitposts on mlops.community ;) Here is a ClearML pipeline integrated with the notorious mlops_this generated by GitHub's Copilot. ClearML is the only open-source tool to manage all your MLOps in a unified and robust platform providing collaborative experiment management, powerful orchestration, easy-to-build data stores, and one-click model deployment.

[MLOps] The Clear SHOW - S02E12 - Goodbye Fig .1 [Sculley15]

Sometimes, even in a field as young and bustling, one has to say goodbye to an old friend. Today we bid adieu to Fig. 1 of D. Sculley et al., AKA "Hidden technical debt in Machine learning systems." Listen to Ariel Biller explaining what's going on and what are we going to use in lieu of Fig. 1

[MLOps] The Clear SHOW - S02E11 - DIY Strikes Back! Building the Model Store!

Ariel extends ClearML's "experiment first" approach towards a "model first" approach - by building a model store. See how easy it is to add metadata to the model artifacts. + Colab notebook (uses the demo server, just run it and see what happens) ClearML is the only open-source tool to manage all your MLOps in a unified and robust platform providing collaborative experiment management, powerful orchestration, easy-to-build data stores, and one-click model deployment.

[MLOPS] From #GTC21: Best Practices in Handling Machine Learning Pipelines on DGX Clusters

Learn how to set up and orchestrate end-to-end ML pipelines, leveraging large DGX clusters. We'll demonstrate how to orchestrate your training and inference workloads on DGX clusters, with optional setup of remote development environments leveraging the multi-instance GPUs on the NVIDIA A100. We'll also show how pipelines can be built to serve both research and deployment workloads, all while leveraging the compute inherent in the DGX cluster.

[MLOPS] From #GTC21: How to Supercharge Your Team's Productivity with MLOps

Learn how to structure a data scientist-first orchestration setup that allows your DS team to self-manage their allocated NVIDIA GPU clusters, without needing continuous hand-holding from DevOps/IT. We'll demonstrate this setup while using NVIDIA Clara Train SDK to walk through best practices in orchestration, experiment management, and data operations and pipelining. While examples will be health-care-focused, the concepts demonstrated are agnostic to any ML/DL use case in any industry.

[MLOPS] From #GTC21: Workshop - Demonstrating an End-to-End Pipeline for ML/DL Leveraging GPUs

Learn how to take models from research into deployment in an efficient and scalable manner. We'll demonstrate workflows and methodologies so that your data science team can make the most of their NVIDIA hardware systems and software tools (including TRITON!).