soopervisor
couler
Our great sponsors
soopervisor | couler | |
---|---|---|
3 | 1 | |
42 | 885 | |
- | 1.6% | |
5.0 | 5.2 | |
3 months ago | 11 days ago | |
Python | Python | |
Apache License 2.0 | Apache License 2.0 |
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
soopervisor
- Ploomber Cloud - Parametrizing and running notebooks in the cloud in parallel
-
Show HN: Ploomber Cloud (YC W22) – run notebooks at scale without infrastructure
Hi, we’re Ido & Eduardo, the founders of Ploomber. We’re launching Ploomber Cloud today, a service that allows data scientists to scale their work from their laptops to the cloud.
Our open-source users (https://github.com/ploomber/ploomber) usually start their work on their laptops; however, often, their local environment falls short, and they need more resources. Typical use cases run out of memory or optimize models to squeeze out the best performance. Ploomber Cloud eases this transition by allowing users to quickly move their existing projects into the cloud without extra configurations. Furthermore, users can request custom resources for specific tasks (vCPUs, GPUs, RAM).
Both of us experienced this challenge firsthand. Analysis usually starts in a local notebook or script, and whenever we wanted to run our code on a larger infrastructure we had to refactor the code (i.e. rewrite our notebooks using Kubeflow’s SDK) and add a bunch of cloud configurations. Ploomber Cloud is a lot simpler, if your notebook or script runs locally, you can run it in the cloud with no code changes and no extra configuration. Furthermore, you can go back and forth between your local/interactive environment and the cloud.
We built Ploomber Cloud on top of AWS. Users only need to declare their dependencies via a requirements.txt file, and Ploomber Cloud will take care of making the Docker image and storing it on ECR. Part of this implementation is open-source and available at: https://github.com/ploomber/soopervisor
Once the Docker image is ready, we spin up EC2 instances to run the user’s pipeline distributively (for example, to run hundreds of ML experiments in parallel) and store the results in S3. Users can monitor execution through the logs and download artifacts. If source code hasn’t changed for a given pipeline task, we use cached artifacts and skip redundant computations, severely cutting each run's cost, especially for pipelines that require GPUs.
Users can sign up to Ploomber Cloud for free and get started quickly. We made a significant effort to simplify the experience (https://docs.ploomber.io/en/latest/cloud/index.html). There are three plans (https://ploomber.io/pricing/): the first is the Community plan, which is free with limited computing. The Teams plan has a flat $50 monthly and usage-based billing, and the Enterprise plan includes SLAs and custom pricing.
We’re thrilled to share Ploomber Cloud with you! So if you’re a data scientist who has experienced these endless cycles of getting a machine and going through an ops team, an ML engineer who helps data scientists scale their work, or you have any feedback, please share your thoughts! We love discussing these problems since exchanging ideas sparks exciting discussions and brings our attention to issues we haven’t considered before!
- Export Ploomber pipelines to Kubernetes (Argo), Airflow, AWS Batch, SLURM, and Kubeflow
couler
-
(Not) to Write a Pipeline
author seems to be describing the kind of patterns you might make with https://argoproj.github.io/argo-workflows/ . or see for example https://github.com/couler-proj/couler , which is an sdk for describing tasks that may be submitted to different workflow engines on the backend.
it's a little confusing to me that the author seems to object to "pipelines" and then equate them with messaging-queues. for me at least, "pipeline" vs "workflow-engine" vs "scheduler" are all basically synonyms in this context. those things may or may not be implemented with a message-queue for persistence, but the persistence layer itself is usually below the level of abstraction that $current_problem is really concerned with. like the author says, eventually you have to track state/timestamps/logs, but you get that from the beginning if you start with a workflow engine.
i agree with author that message-queues should not be a knee-jerk response to most problems because the LoE for edge-cases/observability/monitoring is huge. (maybe reach for a queue only if you may actually overwhelm whatever the "scheduler" can handle.) but don't build the scheduler from scratch either.. use argowf, kubeflow, or a more opinionated framework like airflow, mlflow, databricks, aws lamda or step-functions. all/any of these should have config or api that's robust enough to express rate-limit/retry stuff. almost any of these choices has better observability out-of-the-box than you can easily get from a queue. but most importantly.. they provide idioms for handling failure that data-science folks and junior devs can work with. the right way to structure code is just much more clear and things like structuring messages/events, subclassing workers, repeating/retrying tasks, is just harder to mess up.
What are some alternatives?
kfserving - Standardized Serverless ML Inference Platform on Kubernetes [Moved to: https://github.com/kserve/kserve]
community - Information about the Kubeflow community including proposals and governance information.
pipelines - Machine Learning Pipelines for Kubeflow
argo - Workflow Engine for Kubernetes
elyra - Elyra extends JupyterLab with an AI centric approach.
awesome-argo - A curated list of awesome projects and resources related to Argo (a CNCF graduated project)
hera - Hera is an Argo Python SDK. Hera aims to make construction and submission of various Argo Project resources easy and accessible to everyone! Hera abstracts away low-level setup details while still maintaining a consistent vocabulary with Argo. ⭐️ Remember to star!
kserve - Standardized Serverless ML Inference Platform on Kubernetes
ploomber - The fastest ⚡️ way to build data pipelines. Develop iteratively, deploy anywhere. ☁️
argo-workflows-aws-plugin - Argo Workflows Executor Plugin for AWS Services, e.g. SageMaker Pipelines, Glue, etc.