CausalBind / data /scripts /train.sh
introvoyz045's picture
Migrated from GitHub
d61c939 verified
Raw History Blame Contribute Delete
2.25 kB
#!/usr/bin/env bash
# Train a CausalBind variant on 4 GPUs (50 epochs, CASF validation).
# Usage: bash scripts/train.sh <sp|lr|emb|atom|hypseek> [save_root] [seed]
set -euo pipefail
variant="$1"
save_root="${2:-./save}"
seed="${3:-1}"
REPO_DIR="$(cd "$(dirname "$0")/.." && pwd)"
source "${REPO_DIR}/scripts/variant_config.sh" "${variant}"
data_path="${REPO_DIR}/data"
save_name="causalbind_${variant}_seed${seed}"
save_dir="${save_root}/${save_name}/savedir"
tmp_save_dir="${save_root}/${save_name}/tmp_save_dir"
tsb_dir="${save_root}/${save_name}/tsb_dir"
mkdir -p "${save_dir}" "${tmp_save_dir}" "${tsb_dir}" "${save_root}/train_log"
n_gpu="${N_GPU:-4}"
MASTER_PORT="${MASTER_PORT:-10344}"
finetune_mol_model="${REPO_DIR}/pretrain/mol_pre_no_h_220816.pt"
finetune_pocket_model="${REPO_DIR}/pretrain/pocket_pre_220816.pt"
export OMP_NUM_THREADS=1
export NCCL_ASYNC_ERROR_HANDLING=1
UNICORE_TRAIN=$(which unicore-train)
torchrun --nproc-per-node=${n_gpu} --master-port=${MASTER_PORT} \
${UNICORE_TRAIN} ${data_path} \
--user-dir "${REPO_DIR}/unimol" \
--task train_task \
--arch ${arch} \
--loss three_hybrid_loss \
--train-subset train --valid-subset valid \
--valid-set CASF \
--num-workers 0 --ddp-backend=c10d \
--max-pocket-atoms 256 \
${model_flags} \
--optimizer adam --adam-betas "(0.9, 0.999)" --adam-eps 1e-8 --clip-norm 1.0 \
--lr-scheduler polynomial_decay --lr 1e-4 --warmup-ratio 0.06 --max-epoch 50 \
--batch-size 24 --batch-size-valid 32 \
--fp16 --fp16-init-scale 4 --fp16-scale-window 256 \
--update-freq 1 --seed ${seed} \
--tensorboard-logdir ${tsb_dir} \
--log-interval 100 --log-format simple \
--validate-interval 1 \
--all-gather-list-size 2048000 \
--save-interval 1 \
--save-dir ${save_dir} --tmp-save-dir ${tmp_save_dir} \
--keep-best-checkpoints 8 --keep-last-epochs 50 \
--find-unused-parameters \
--finetune-pocket-model ${finetune_pocket_model} \
--finetune-mol-model ${finetune_mol_model} \
--max-lignum 16 \
--learn-curv \
--protein-similarity-thres 1.0 \
--best-checkpoint-metric valid_bedroc --maximize-best-checkpoint-metric \
2>&1 | tee "${save_root}/train_log/train_log_${save_name}.txt"