diff options
Diffstat (limited to 'scripts/run_ladder.sh')
| -rwxr-xr-x | scripts/run_ladder.sh | 10 |
1 files changed, 9 insertions, 1 deletions
diff --git a/scripts/run_ladder.sh b/scripts/run_ladder.sh index 65b43a9..17219bd 100755 --- a/scripts/run_ladder.sh +++ b/scripts/run_ladder.sh @@ -30,7 +30,15 @@ EXTRA="--set ${TOKENS:+tokens=$TOKENS} ${SET:-}" echo "== zbp-scaling ladder: sizes=[$SIZES] arms=[$ARMS] nproc=$NPROC data=$DATA out=$OUT ${TOKENS:+tokens=$TOKENS}" [ "$NPROC" -ge 1 ] || { echo "no GPUs detected; set NPROC"; exit 1; } -python -c "import zbp_scaling" 2>/dev/null || pip install -e . -q +# --- environment bootstrap: current env first; VENV=1 creates ./.venv (fresh nodes) --- +PY=${PY:-python3} +if [ "${VENV:-0}" = 1 ]; then + [ -d .venv ] || "$PY" -m venv .venv + . .venv/bin/activate +fi +python -c "import zbp_scaling" 2>/dev/null || pip install -e . -q || { + echo "pip install failed; on managed nodes run with VENV=1 (creates ./.venv) or install into your conda env"; exit 1; } +python scripts/env_check.py || exit 1 if [ ! -f "$DATA/train.bin" ]; then echo "== preparing data ($PREP_TOKENS tokens -> $DATA) — run this on the training node" |