Skip to content

README.md

Latest commit

 

History

History
34 lines (27 loc) · 981 Bytes

README.md

File metadata and controls

34 lines (27 loc) · 981 Bytes

deployable-classifier python

A modular, pytorch-based CNN classifier for CIFAR-10, packaged into a RESTful API service using FastAPI.

TODO:

  • Train and test basic model.
  • Write the api-based inference script.
  • Test how the model handles OOD images.
  • Add batching or queueing to inference script.
  • Add detection for performance degradation/increasing error.
  • Test robustness to adversarial inputs.

Quickstart

  1. Install dependencies:
uv pip install -r requirements.txt
  1. Train model:
accelerate launch -m src.train
  1. Start the API server:
uvicorn src.api:app --reload &
  1. Run test inference scripts:
python -m src.api_infer_serial --config=src/configs/common.py # for serial mode
python -m src.api_infer_batched --config=src/configs/common.py # for batched mode