#!/bin/bash
#SBATCH --job-name=fm_train
#SBATCH --partition=workq
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=10
#SBATCH --time=24:00:00
#SBATCH --output=outputs/logs/train_%j.log
#SBATCH --error=outputs/logs/train_%j.log
#
# Train/eval the Tx + EHD models (scripts/11) on a dedicated compute node.
# Extra args after the script name are forwarded to 11_train_eval.py, e.g.
#   sbatch scripts/train_eval.sbatch --max-files 3 --max-iter 40 --train-cap 200000   (smoke)
#   sbatch scripts/train_eval.sbatch                                                   (full)

cd /home/tanpv/Progs/claude/FM || exit 1
NT=${SLURM_CPUS_PER_TASK:-16}
export OMP_NUM_THREADS=$NT OPENBLAS_NUM_THREADS=$NT MKL_NUM_THREADS=$NT NUMEXPR_NUM_THREADS=$NT
PY=/home/hapt/miniconda3/bin/python3
echo "=== fm_train on $(hostname) | cpus=$NT | $(date) ==="
echo "args: $*"
$PY scripts/11_train_eval.py "$@"
rc=$?
echo "=== exit $rc | $(date) ==="
exit $rc