NNCF scaling proposal based on the PyTorch Timm project

This is just an experimental project that shows the results of applying NNCF quantization to a subset of models from Timm.

Installation

• Clone this repo
• Install dependencies
  • pip install -r requirements.txt
  • Clone OpenVINO Model Analyzer in the working directory.

Running the script

• Just execute the script with the parameter that specifies a directory where the output models will be bumped:

nncf_timm.py ./models

• See the reults in the log.txt file or in console log