File size: 3,434 Bytes
7718ac6 | 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 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 | #!/usr/bin/env bash
# =============================================================================
# GSVC one-click launcher.
#
# Unpacks the bundled Python environment (first run only), wires up all model
# caches so nothing is downloaded, and runs the acceptance harness end-to-end.
#
# Usage:
# ./run.sh # run full acceptance over all 5 datasets
# ./run.sh --clips uvg_beauty,mcl_src01 # a subset of clips
#
# Requirements on the host: only a CUDA-capable GPU + driver and `bash`,
# `tar`. Everything else (Python, PyTorch, ffmpeg, all model weights) ships
# inside this bundle. No internet access is required.
# =============================================================================
set -euo pipefail
ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
ENV_DIR="${ROOT}/env"
ENV_TARBALL="${ROOT}/gsvc_env.tar.gz"
CODE_DIR="${ROOT}/code"
# ---------------------------------------------------------------------------
# 0. Extract the code and data archives (first run only).
# ---------------------------------------------------------------------------
if [ ! -d "${CODE_DIR}" ] && [ -f "${ROOT}/code.tar.gz" ]; then
echo "[run] first-time setup: extracting code..."
tar -xzf "${ROOT}/code.tar.gz" -C "${ROOT}"
fi
if [ ! -d "${ROOT}/data" ] && [ -f "${ROOT}/data.tar.gz" ]; then
echo "[run] first-time setup: extracting test data..."
tar -xzf "${ROOT}/data.tar.gz" -C "${ROOT}"
fi
# ---------------------------------------------------------------------------
# 1. Unpack the relocatable conda environment (idempotent).
# ---------------------------------------------------------------------------
if [ ! -x "${ENV_DIR}/bin/python" ]; then
echo "[run] first-time setup: unpacking Python environment (~19 GB, a few minutes)..."
mkdir -p "${ENV_DIR}"
tar -xzf "${ENV_TARBALL}" -C "${ENV_DIR}"
# conda-unpack rewrites absolute paths baked into the env at pack time.
if [ -x "${ENV_DIR}/bin/conda-unpack" ]; then
"${ENV_DIR}/bin/conda-unpack"
fi
echo "[run] environment ready."
fi
PY="${ENV_DIR}/bin/python"
# ---------------------------------------------------------------------------
# 2. Point every cache at the bundled copies -> fully offline.
# ---------------------------------------------------------------------------
export HF_HOME="${ROOT}/hf_cache"
export HUGGINGFACE_HUB_CACHE="${ROOT}/hf_cache/hub"
export TRANSFORMERS_CACHE="${ROOT}/hf_cache/hub"
export TORCH_HOME="${ROOT}/torch_cache"
export HF_HUB_OFFLINE=1
export TRANSFORMERS_OFFLINE=1
export GSVC_LTX_INFER_DTYPE=bf16
export PYTORCH_ALLOC_CONF=expandable_segments:True
# The gsvc code + vendored trainer are used in place (editable-style) via
# PYTHONPATH, so the packed env does not need them installed.
export PYTHONPATH="${CODE_DIR}/src:${CODE_DIR}/external/LTX-Video-Trainer/src:${CODE_DIR}/external/LTX-Video:${PYTHONPATH:-}"
# The gsvc runtime resolves its worker interpreter as <code>/.venv-ltx/bin/python.
# Expose the unpacked env there so inference + metrics subprocesses find it.
if [ ! -e "${CODE_DIR}/.venv-ltx" ]; then
ln -s "${ENV_DIR}" "${CODE_DIR}/.venv-ltx"
fi
# ---------------------------------------------------------------------------
# 3. Run the acceptance harness.
# ---------------------------------------------------------------------------
echo "[run] starting GSVC acceptance..."
cd "${CODE_DIR}"
exec "${PY}" scripts/gsvc_acceptance.py "$@"
|