Show HN: Valohai – MLOps and Record Keeping for ML https://ift.tt/2rwbfIa

Show HN: Valohai – MLOps and Record Keeping for ML Hi HN, we've been here before [1] but the product has changed quite. In the time of Kubeflow, MLFlow, and Metaflow: Valohai is a managed fullfledged platform for ML. Unlike the others, it doesn't require you to set up and maintain it yourself. And unlike the rest, it doesn't require your code to import a library that only works for python. We'd love to hear your feedback and thoughts! Valohai in a few bullet-points: 1) MLOps - Runs on your AWS, GCP, Azure, on-prem - Automatically starts and shuts down your servers once your code is done (no need to manually track and shut down e.g. EC2 instances) - Launch parallel hyperparameter sweeps (e.g. 100 TPUs at once doing 100 combinations of hyperparams) - Works with code on localhost, gitlab/github/bitbucket and jupyter notebooks (all is run in your cloud/serverfarm, as opposed to localhost) - Nifty features such as async jupyter execs (change code/data while previous code is still running in the cloud) 2) Record keeping - Automatically tracks the code commit, data, hyperparameters, libraries, runtime hardware, etc (technically a reference to commit hash, docker container, data in bucket, hyperparams and all logs) - One-click redoing of experiment with same params above - Data traceability view (see all scripts, data etc that the model went through) - Everything is built around teams and projects: Share projects with teams and follow progress in real-time. Don't lose what others did and roll back to what you did last Friday at the click of a button. 3) Fully managed service - Run all job workers in your own AWS, GCP, Azure, on-prem - Use the Valohai orchestrator (that launches the above workers) in your cloud, our cloud or on-prem - Everything is fully set up, managed, updated and maintained by us Try it out yourself: Create a free account, run code, experiment, play at https://valohai.com/ What are your thoughts? [1] https://ift.tt/2ncWDsl December 9, 2019 at 01:11PM