Custom Model Hosting (BYOM)
Upload custom ONNX, TensorFlow, PyTorch, or HuggingFace models for managed inference
Why do anything?
Managing model infrastructure is complex. Custom models need deployment without DevOps overhead.
Why now?
ML teams have trained models but lack deployment expertise. Managed inference reduces friction.
Why this feature?
BYOM accepts ONNX, TensorFlow, PyTorch, and HuggingFace models. Mixpeek handles deployment, scaling, and inference.
How It Works
BYOM provides managed model deployment on Mixpeek infrastructure.
1
Upload
Upload model via presigned URL or streaming
2
Validation
Validate format and security scan
3
Deployment
Deploy to Ray cluster with configured resources
4
Integration
Enable as feature extractor for collections
Why This Approach
Managed deployment eliminates infrastructure complexity. Ray provides scalable inference.
Where This Is Used
Integration
plugin = client.plugins.byom.upload(model_path="model.onnx", config={...})