scripts/run_predictor_v4_training.sh

#!/usr/bin/env bash
set -euo pipefail

PROJECT_ROOT="/mnt/s3files/s3-us-west2-default/zoubin/cz/projects/Self-Forcing"
CONFIG_PATH="${CONFIG_PATH:-${PROJECT_ROOT}/configs/predictor_v4.yaml}"
NPROC_PER_NODE="${NPROC_PER_NODE:-8}"
MASTER_PORT="${MASTER_PORT:-29641}"
LOG_DIR="${LOG_DIR:-/mnt/local_nvme/zoubin/cz/self_forcing_predictor_v4_training_1000_seed0}"
PYTHON_BIN="${PYTHON_BIN:-/mnt/s3files/s3-us-west2-default/zoubin/cz/envs/self_forcing/bin/python}"

cd "${PROJECT_ROOT}"
mkdir -p "${LOG_DIR}"

"${PYTHON_BIN}" -m torch.distributed.run \
  --standalone \
  --nproc_per_node="${NPROC_PER_NODE}" \
  --master_port="${MASTER_PORT}" \
  train.py \
  --config_path "${CONFIG_PATH}" \
  --disable-wandb \
  2>&1 | tee -a "${LOG_DIR}/launcher.log"