Getting started
Serverless GPU inference for ML models.
Pipeline is a python API that allows you to execute your local code anywhere.
It's primarily designed for ML engineers that want to run their code remotely without having to do any DevOps. The library gives you the tools to create a compute graph that can be run at scale without needing setup containers or kubernetes. Everything is done through python (more languages are in the works).
There are 3 primary ways to use code made in Pipeline:
- Run your code on PipelineCloud
- Deploy on your own cloud account with a Licence for PipelineStack (coming soon)
- Auto create dockerised APIs containers to run in your custom Kubernetes infra, or as standalone containers
Key features
- Run on serverless GPU in seconds
- Run code remotely
- Automated scaling
- Reduced cold start
- No DevOps experience required
New to Pipeline?
If you're ready to dive in and learn Pipeline, try out some of the tutorials where you'll create some fully-fledged Pipeline Cloud deployments, or read on for a quick sample of Pipeline.
Quick Start: Hello Pipeline
Install pipeline
with
pip install -U pipeline-ai
See the installation guide for more detailed instructions.
Run a basic pipeline
Import Pipeline
, pipeline_function
and decorate your Python function using the @pipeline_function decorator.
from pipeline import Pipeline, pipeline_function
@pipeline_function
def hello_pipeline() -> str:
return "Hello Pipeline"
Configure your pipeline to use your Python functions through the Pipeline
context manager:
with Pipeline("hello-pipeline") as pipeline_builder:
greeting = hello_pipeline()
pipeline_builder.output(greeting)
Then get
the pipeline and run it:
hello_pipeline = Pipeline.get_pipeline("hello-pipeline")
print(hello_pipeline.run())
And there you go, you're function is now being executed within a pipeline!
Next Steps
Try our cloud quickstart guide to learn about how to run pipelines in the cloud on enterprise GPUs. Then check out our tutorials for guides on how to deploy your own custom pipelines, environments and integrate with third party services. Go deeper and learn about object concepts in the Python API section.
Updated about 19 hours ago