File size: 2,243 Bytes
62b8d90 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 | #!/usr/bin/env bash
# Layout ablations (vbe/ablate.py) of one checkpoint, one process per GPU per benchmark,
# then one scores.json per (variant, budget).
#
# JOBS="rho32:4096:16384,sel_merge:1024" bash scripts/run_ablate.sh
#
# Re-running the same command resumes. Needs the baseline run (scripts/run_eval.sh) of
# the same checkpoint: identical (question, budget) inputs are copied from it.
set -euo pipefail
cd "$(dirname "$0")/.."
MODEL=${MODEL:-/root/OnlineLong_ft/work_dirs/phase3_qbase_full_hybrid_lr1e-5}
BASELINE=${BASELINE:-results/phase3_qbase_full_hybrid_lr1e-5}
OUT_ROOT=${OUT_ROOT:-results/ablate}
JOBS=${JOBS:?JOBS=variant:budget[:budget...],...}
GPUS=${GPUS:-$(nvidia-smi --query-gpu=index --format=csv,noheader | paste -sd,)}
BENCHMARKS=${BENCHMARKS:-"videomme mlvu longvideobench"}
DATA_ROOT=${DATA_ROOT:-/root/benchmarks/eval_root}
OL_SRC=${OL_SRC:-third_party/onlinelong_phase3}
NUM_WORKERS=${NUM_WORKERS:-6}
EXTRA=${EXTRA:-} # e.g. "--limit 2 --no_reuse"
PY=${PY:-/miniconda/envs/ml/bin/python}
export PATH="$(dirname "${PY}"):${PATH}"
export PYTHONPATH="${OL_SRC}:.${PYTHONPATH:+:${PYTHONPATH}}"
IFS=, read -ra GPU_LIST <<< "${GPUS}"
N=${#GPU_LIST[@]}
LOGS=${OUT_ROOT}/logs
mkdir -p "${LOGS}"
echo "[run_ablate] model=${MODEL} jobs=${JOBS} gpus=${GPUS} benchmarks=${BENCHMARKS} -> ${OUT_ROOT}"
for BENCH in ${BENCHMARKS}; do
pids=()
for i in "${!GPU_LIST[@]}"; do
CUDA_VISIBLE_DEVICES=${GPU_LIST[$i]} "${PY}" -m vbe.ablate \
--model_path "${MODEL}" --benchmark "${BENCH}" --data_root "${DATA_ROOT}" \
--out_root "${OUT_ROOT}" --baseline_run "${BASELINE}" --jobs "${JOBS}" \
--shard_id "${i}" --num_shards "${N}" --num_workers "${NUM_WORKERS}" ${EXTRA} \
> "${LOGS}/${BENCH}_shard${i}of${N}.log" 2>&1 &
pids+=($!)
done
fail=0
for p in "${pids[@]}"; do wait "${p}" || fail=1; done
if [ "${fail}" -ne 0 ]; then
echo "[run_ablate] ${BENCH}: a shard failed, see ${LOGS}/${BENCH}_*.log" >&2
exit 1
fi
echo "[run_ablate] ${BENCH} done"
done
for d in "${OUT_ROOT}"/*_b*/; do
"${PY}" -m vbe.score "${d%/}" "${DATA_ROOT}" > /dev/null
done
echo "[run_ablate] scored $(ls -d "${OUT_ROOT}"/*_b*/ | wc -l) runs"
|