Slurm scripts for downstream tasks

[zhan8855]

Hi, thank you for your awesome work!

Could you please release the exact script to reproduce downstream results? i.e. Could you please fill out this part in run_genomics_benchmark.sh and run_nucleotide_transformer.sh, for both Caduceus-Ph and Caduceus-Ps?

Expected args:

• CONFIG_PATH
• PRETRAINED_PATH
• DISPLAY_NAME
• MODEL
• MODEL_NAME
• CONJOIN_TRAIN_DECODER
• CONJOIN_TEST
• TASK
• LR
• BATCH_SIZE
• RC_AUG

Thank you very much in advance!

[yair-schiff]

The exact values of these parameters are set in the wrapper_ scripts for each of the downstream tasks:

• https://github.com/kuleshov-group/caduceus/blob/main/slurm_scripts/wrapper_run_genomics.sh
• https://github.com/kuleshov-group/caduceus/blob/main/slurm_scripts/wrapper_run_nucleotide_transformer.sh

The appendix in the paper has additional details about hyperparameters such as BATCH_SIZE and LR for each subtask within a benchmark

[zhan8855]

Thank you so much!!