| #PBS -N obf_ae_mistral | |
| #PBS -l select=1:ncpus=16:mem=110G:ngpus=1 | |
| #PBS -l walltime=12:00:00 | |
| #PBS -j oe | |
| #PBS -k oe | |
| #PBS -o safetynet/logs/obf_sim_output.txt | |
| #PBS -P personal-maheep00 | |
| #PBS -q normal | |
| # Go to the directory where the job was submitted | |
| cd $PBS_O_WORKDIR | |
| # Initialize conda | |
| source ~/.bashrc | |
| # Activate the conda environment | |
| conda activate safebymi | |
| # Set environment variables | |
| export PYTHONPATH=$PYTHONPATH:safetynet | |
| export HF_HOME=safetynet/safetynet/huggingface | |
| # Define models to train | |
| MODELS=('mistral') # 'llama2' llama3' 'gemma' 'qwen' 'mistral') # as needed | |
| # Run training for each model | |
| for MODEL in "${MODELS[@]}"; do | |
| python -m src.training.obfuscation --model ${MODEL} --model_type "obfuscated_ae" --dataset spylab \ | |
| > safetynet/logs/${MODEL}/obf_sim.log 2>&1 | |
| done | |