#!/bin/bash -vx echo "$SM_HPS" data_dir=$SM_CHANNEL_GENETIC fasta_path=$SM_CHANNEL_FASTA msa_path=$SM_CHANNEL_MSA openfold_checkpoint_path=$SM_CHANNEL_PARAM echo "data_dir is $data_dir" echo "fasta_path is $fasta_path" echo "msa_path is $msa_path" echo "openfold_checkpoint_path is $openfold_checkpoint_path" ls $data_dir ls $fasta_path ls $openfold_checkpoint_path WDIR=$SM_MODEL_DIR # output model directory, used as the working directory. Files in this dir will be package up in a model.tar.gz and upload to S3 at the end of the job if [ ! -f ${msa_path}/model.tar.gz ]; then exit 1; fi if [ -f ${msa_path}/model.tar.gz ]; then tar xfzv ${msa_path}/model.tar.gz -C $WDIR cd $WDIR/msas/ mv * ../ fi python /opt/openfold/run_pretrained_openfold.py \ ${fasta_path} \ ${data_dir}/pdb_mmcif/mmcif_files/ \ --model_device "cuda:0" \ --jackhmmer_binary_path /opt/conda/bin/jackhmmer \ --hhblits_binary_path /opt/conda/bin/hhblits \ --hhsearch_binary_path /opt/conda/bin/hhsearch \ --kalign_binary_path /opt/conda/bin/kalign \ --config_preset "model_1_ptm" \ --openfold_checkpoint_path ${openfold_checkpoint_path}/finetuning_ptm_2.pt \ --preset $SM_HP_DB_PRESET \ --output_dir $SM_MODEL_DIR \ --use_precomputed_alignments $SM_MODEL_DIR/