code/eval_audioldm2_test_loss.sbat

#!/bin/bash
#SBATCH --job-name=al2_test_loss_sweep
#SBATCH --mail-type=FAIL,END
#SBATCH --mail-user=dingqy@umich.edu
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=1
#SBATCH --cpus-per-task=4
#SBATCH --mem-per-cpu=11GB
#SBATCH --time=4:00:00
#SBATCH --account=ahowens1
#SBATCH --partition=spgpu
#SBATCH --qos=foe
#SBATCH --reservation=spgpu_foe_nodes
#SBATCH --gres=gpu:a40:1
#SBATCH --output=/nfs/turbo/coe-ahowens-nobackup/dingqy/al2_test_loss_sweep-%x-%j.log
#
# audioldm2 test-loss sweep across all production ckpts (concat + controlnet).
# Sharded into 4 single-GPU jobs:
#   SHARD=1 → controlnet 500 - 12000 (9 ckpts)
#   SHARD=2 → controlnet 14000 - 30000 (9 ckpts)
#   SHARD=3 → concat 2000 - 22000 (11 ckpts)
#   SHARD=4 → concat 24000 - 38600 (19 ckpts)
#
# Submit:
#   for s in 1 2 3 4; do SHARD=$s sbatch eval_audioldm2_test_loss.sbat; done
set -eu

source /nfs/turbo/coe-ahowens-nobackup/dingqy/miniforge3/etc/profile.d/conda.sh
source /home/dingqy/.hf_token
export HF_CACHE=/nfs/turbo/coe-ahowens-nobackup/dingqy/.cache/huggingface

conda activate new_audioldm

: "${SHARD:=1}"

build_cn()     { for s in "$@"; do printf "audioldm2_bg2fg_controlnet_rebalance/step_%08d\n" $s; done; }
build_concat() { for s in "$@"; do printf "audioldm2_bg2fg_rebalance/step_%08d\n"            $s; done; }

case "$SHARD" in
    1) CKPTS=$(build_cn 500 1000 1500 2000 4000 6000 8000 10000 12000) ;;
    2) CKPTS=$(build_cn 14000 16000 18000 20000 22000 24000 26000 28000 30000) ;;
    3) CKPTS=$(build_concat 2000 4000 6000 8000 10000 12000 14000 16000 18000 20000 22000) ;;
    4) CKPTS=$(build_concat 24000 26000 28000 30000 32000 34000 36000 36200 36400 36600 36800 37000 37200 37400 37600 37800 38000 38200 38600) ;;
    *) echo "unknown SHARD=$SHARD (use 1/2/3/4)"; exit 1 ;;
esac

OUT=/nfs/turbo/coe-ahowens-nobackup/dingqy/al2_test_loss_shard${SHARD}.json
echo "shard=$SHARD  out=$OUT"
echo "$CKPTS"
echo ""

python /nfs/turbo/coe-ahowens-nobackup/dingqy/eval_audioldm2_test_loss.py \
    --ckpts $CKPTS \
    --out "$OUT" \
    --batch-size 8 --num-workers 4

echo ""
echo "=== shard $SHARD final ==="
python -c "
import json
r = json.load(open('$OUT'))
for k, v in sorted(r.items(), key=lambda x: x[1]['test_loss']):
    print(f'  {v[\"test_loss\"]:.5f}  ({v[\"arch\"]:>10})  {k}')
"