ray

Ray distributed computing framework. Use for scaling ML.

7 stars

byG1Joshi

View on GitHub Installation ↓

Best use case

ray is best used when you need a repeatable AI agent workflow instead of a one-off prompt.

Ray distributed computing framework. Use for scaling ML.

Teams using ray should expect a more consistent output, faster repeated execution, less prompt rewriting.

When to use this skill

You want a reusable workflow that can be run more than once with consistent structure.

When not to use this skill

You only need a quick one-off answer and do not need a reusable workflow.
You cannot install or maintain the underlying files, dependencies, or repository context.

Installation

Claude Code / Cursor / Codex

$curl -o ~/.claude/skills/ray/SKILL.md --create-dirs "https://raw.githubusercontent.com/G1Joshi/Agent-Skills/main/skills/ai-ml/ray/SKILL.md"

Manual Installation

Download SKILL.md from GitHub
Place it in .claude/skills/ray/SKILL.md inside your project
Restart your AI agent — it will auto-discover the skill

How ray Compares

Feature / Agent	ray	Standard Approach
Platform Support	Not specified	Limited / Varies
Context Awareness	High	Baseline
Installation Complexity	Unknown	N/A

Frequently Asked Questions

What does this skill do?

Ray distributed computing framework. Use for scaling ML.

Where can I find the source code?

You can find the source code on GitHub using the link provided at the top of the page.

SKILL.md Source

# Ray

Ray is the compute layer for AI. It powers ChatGPT training and massive scale workloads. v3.0 (2025) improves **efficiency** and adds an **MCP Server** for agents.

## When to Use

- **Distributed Training**: Scaling PyTorch across 100 GPUs.
- **Ray Serve**: Serving LLMs with high throughput (vLLM integration).
- **Hyperparameter Tuning**: Ray Tune is the industry standard.

## Core Concepts

### Actors & Tasks

- **Task**: Stateless function (like Lambda).
- **Actor**: Stateful class (like a microservice).

### Object Store

Shared memory across the cluster means zero-copy data sharing.

## Best Practices (2025)

**Do**:

- **Use `ray.data`**: For streaming massive datasets into trainers.
- **Use KubeRay**: The Kubernetes operator for managing Ray clusters.
- **Use Ray Serve**: It supports "Model Composition" (chaining models).

**Don't**:

- **Don't use for simple scripts**: The overhead of starting a Ray cluster is 5-10s.

## References

- [Ray Documentation](https://docs.ray.io/)