https://github.com/mcaimi/kserve-diffusers-demo
diffusers-based model server for KServe
https://github.com/mcaimi/kserve-diffusers-demo
Last synced: about 2 months ago
JSON representation
diffusers-based model server for KServe
- Host: GitHub
- URL: https://github.com/mcaimi/kserve-diffusers-demo
- Owner: mcaimi
- License: gpl-3.0
- Created: 2024-06-24T13:00:34.000Z (about 2 years ago)
- Default Branch: main
- Last Pushed: 2024-12-02T10:45:15.000Z (over 1 year ago)
- Last Synced: 2025-03-06T10:22:03.385Z (over 1 year ago)
- Language: Python
- Homepage:
- Size: 31.3 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# KServe Compatible Demo Model Server
This is a demo model server that can run inside Openshift AI as a KServe model server.
It exposes a way to perform inference with a Stable Diffusion compatible model in the backend.
## Parameters
The server expects a JSON-encoded payload to start inference:
```json
// example payload:
{
"instances": [
{
"prompt": "photo of the beach",
"negative_prompt": "ugly, deformed, bad anatomy",
"num_inference_steps": 20,
"scheduler": "DPM++ SDE",
}
]
}
```
## Known Parameters:
- "prompt": the stable diffusion positive prompt
- "negative_prompt": put here all negative embeddings
- "num_inference_steps": number of generation steps to run during inference
- "width" and "height": size of the generated image
- "guidance_scale": the guidance scale value to feed to the neural network
- "seed": if specified, use this value as the generation seed.
- "scheduler": Noise Scheduler to use during generation (DPM++ 2M, DPM++ SDE, Euler and Euler a, Heun and LMS)