Skip to content

README.md

Latest commit

 

History

History
126 lines (89 loc) · 2.78 KB

README.md

File metadata and controls

126 lines (89 loc) · 2.78 KB
Raw

01_network_volumes

Network volume example that shares model cache and generated images between GPU and CPU workers.

Overview

The GPU worker generates images with Stable Diffusion and writes them to a Runpod network volume. The CPU worker lists or serves those images from the same volume.

Quick Start

1. Install Dependencies

uv sync

2. Authenticate

uv run flash login

Or create a .env file with RUNPOD_API_KEY=your_api_key_here.

3. Run Locally

uv run flash dev

Server starts at http://localhost:8888

4. Test the API

Generate an image (GPU worker):

curl -X POST http://localhost:8888/gpu_worker/runsync \
  -H "Content-Type: application/json" \
  -d '{"prompt": "a sunset over mountains"}'

List generated images (CPU worker):

curl http://localhost:8888/images

Get a specific image (CPU worker):

curl http://localhost:8888/images/sd_generated_20240101_120000.png

Visit http://localhost:8888/docs for interactive API documentation.

What You'll Learn

  • How to attach a shared NetworkVolume to GPU and CPU workers
  • How to use the GPU worker to write to the volume
  • How to read and serve files from the CPU worker

Architecture

  • GPU worker: loads Stable Diffusion, caches weights in /runpod-volume/models, writes images to /runpod-volume/generated_images
  • CPU worker: lists images and serves files from /runpod-volume/generated_images

Quick Project Structure

01_network_volumes/
├── gpu_worker.py        # Stable Diffusion worker with @Endpoint
├── cpu_worker.py        # List and serve images with Endpoint routes
├── requirements.txt
└── README.md

API Endpoints

POST /gpu_worker/runsync

GPU worker (QB, class-based @Endpoint). Generates an image and saves it to the shared volume.

Request:

{ "prompt": "a sunset over mountains" }

Response:

{
  "prompt": "string",
  "image_path": "string",
  "timestamp": "string",
  "generation_params": {
    "num_inference_steps": 20,
    "guidance_scale": 7.5,
    "width": 512,
    "height": 512
  },
  "message": "Image generated and saved locally!"
}

GET /images

CPU worker (LB, explicit path). Lists all generated images on the shared volume.

Response:

{ "status": "success", "images": ["sd_generated_20240101_120000.png"] }

GET /images/{file_name}

CPU worker (LB, explicit path). Returns metadata and base64-encoded content for a single image.

Notes

  • The network volume is defined inline in each worker file and attached to both workers.
  • Stable Diffusion weights are cached in the volume so cold starts are faster after the first run.

Deployment

flash build
flash deploy new production
flash deploy send production