Cluster & HPC

We provide SLURM-oriented scripts for cluster workflows.

Pipeline Scripts

Script	Description
complex_and_ts_search_local.sh	Local pipeline template
complex_and_ts_search_cpu.sh	SLURM CPU pipeline template
complex_and_ts_search_gpu.sh	GPU validator-oriented template
validation_goflow.sh	Validate model-generated TS guesses on cluster

Running on SLURM

As with all the other scripts, make sure that to adjust them according to your environment.

sbatch complex_and_ts_search_cpu.sh

Parallel Execution with Hydra

python -m motsart.complex_finder.complex_finder -m \
    hydra/launcher=joblib \
    hydra.launcher.n_jobs=4 \
    env=cluster \
    "env.rxn_num=range(0,32)"

Learning on Cluster

Typical workflow:

1. Push code and data: bash push_to_musica.sh
2. Run base pipeline: sbatch complex_and_ts_search_musica.sh
3. Prepare training/eval data: bash create_fine_tune_dft_data.sh or bash create_preprocess_rtsp_pretrain_data.sh
4. Import model-generated samples: bash fetch_and_push_data_pkl_to_results.sh
5. Validate model-generated guesses: sbatch validation_goflow.sh
6. Compute stats:

python -m motsart.validator.compute_stats \
  --cluster-folder /data/results_cluster \
  --learning-folder /data/results_goflow/finetune_noise_1_TS \
  --validator DFTValidator \
  --output-csv /data/results_goflow/finetune_noise_1_TS/stats_al.csv \
  --cluster-ts-method racer_ts \
  --al-ts-method learning \
  --mode both

For a general walkthrough, see Paper Reproduction Workflow.