Ray

Ray is the distributed-computing framework that underpins much of modern AI infrastructure: Ray Serve for scalable model serving, Ray Train for distributed training, Tune for hyperparameter search, and Data for distributed preprocessing — all on one runtime that scales from a laptop to a cluster. 44k stars, Apache-2.0, 13.7M PyPI downloads/month, commercial backing from Anyscale. It's the default serving layer for many serious deployments (and the engine inside several inference stacks). The cost: genuine distributed-systems ops — clusters, autoscaling, and version pinning are your job.

```bash

pip install "ray[serve]"

```

```python

from ray import serve

@serve.deployment

class Model:

def __call__(self, req): return {"ok": True}

serve.run(Model.bind())

```

Docs: docs.ray.io.