zombie-weights-code / scripts /bootstrap_l4.sh
tony0014's picture
fix(circuitkit): ig_steps=2, n_examples=4, finally free_cuda, expandable_segments
6892422 verified
Raw History Blame Contribute Delete
1.39 kB
#!/bin/bash
# Fallback bootstrap for running the L4 smoke test on a stock PyTorch image
# when the pre-baked dual-venv image is unavailable. Creates the same two
# isolated environments in /tmp and execs the smoke script.
#
# Usage (inside an hf jobs container):
# bash /code/scripts/bootstrap_l4.sh
set -euo pipefail
CORE=/tmp/venv_core
ALIGN=/tmp/venv_aligntune
if ! command -v uv >/dev/null 2>&1; then
pip install -q --break-system-packages uv
fi
uv venv --system-site-packages --seed "$CORE"
"$CORE/bin/pip" install --no-cache-dir \
"transformers>=5.15,<5.22" \
"trl>=0.12,<1.15" \
"peft>=0.18,<1" \
"huggingface-hub>=1.5,<2" \
"transformer-lens==3.8.0" \
"bitsandbytes>=0.41" \
"datasets>=3.0" \
"accelerate>=1.4" \
"sentencepiece" \
"protobuf" \
"scikit-learn" \
"hf_transfer" \
"curatorkit[hf,connectors]" \
"circuitkit==0.2.3" \
"safetune==0.1.18" \
"auditkit==1.1.11"
uv venv --system-site-packages --seed "$ALIGN"
"$ALIGN/bin/pip" install --no-cache-dir \
"torchvision==0.28.0" \
"transformers>=5.15,<5.22" \
"peft>=0.18,<1" \
"datasets>=4.7.0" \
"accelerate>=1.6" \
"trl==1.7.1" \
"torchao==0.18.0" \
"aligntune==0.1.35"
export ALIGNTUNE_BIN="$ALIGN/bin/aligntune"
export ALIGNTUNE_PYTHON="$ALIGN/bin/python"
export PYTORCH_CUDA_ALLOC_CONF="expandable_segments:True"
exec "$CORE/bin/python" /code/scripts/smoke_l4.py "$@"