scripts

Instruction on how to run the scripts

---

• ssh into your Intel Dev Cloud Enviornment
• using tmux is preferred it already installed in your env it makea everything smooth (optional)
• run 'tmux new -t main' this will open a terminal within your dev cloud terminal (optional)
• If not using tmux just simply open your dev cloud terminal
• run srun --pty bash
• run source /opt/intel/oneapi/setvars.sh
• run pip install -r requirements.txt
• run pip install intel_extension_for_pytorch==2.0.110+xpu -f https://developer.intel.com/ipex-whl-stable-xpu (if using xpu enabled device)
• Let's start training run python train.py
• If you using tmux press ctrl+b and then d this will minimise your terminal
• To open again run tmux a -t main this resume your terminal without stopping training
• After training is complete navigate your trained model and run bash run_pruning.sh and provide trained model path when prompted
• This will use Intel Neural Compressor to prune your model you can a lot of parameter by changing values in run_pruning.sh file
• Navigate your pruned model and let's start Quantization.
• run bash run_quantization.sh and provide your pruned model path when prompted.
• This will also use Intel Neural Compressor to quantize your model
• Navigate your final quantized model this might not be the most accurate model but will have the highest throughput.
• Now run bash generate_submission.sh provide model of your choice when prompted and it will infer and save results in current folder with name submission.csv

Screenshot showing how training on Intel Dev Cloud