From cd776cfceb19bff30d39e136aa02f25d9fe65c01 Mon Sep 17 00:00:00 2001
From: songhyeonsu <hunsoo0823@naver.com>
Date: Thu, 7 May 2026 16:32:42 +0900
Subject: [PATCH] Rewrite setup_server.sh for paddle-only container

- Targets the dedicated kr_lp_pgnet container (Ubuntu 24.04 + py3.10 + paddle 3.3.0.dev sm_120 wheel)
- Reuses pre-staged wheel under /workspace/wheels/
- Pins PaddleOCR release/2.7 (PGNet algorithm baseline)
- Adds ppocr import smoke test before downloading pretrain
- Symlinks our dict/config into PaddleOCR tree

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
---
 scripts/setup_server.sh | 71 ++++++++++++++++++++++++++++++-----------
 1 file changed, 52 insertions(+), 19 deletions(-)

diff --git a/scripts/setup_server.sh b/scripts/setup_server.sh
index f831689..76e9887 100755
--- a/scripts/setup_server.sh
+++ b/scripts/setup_server.sh
@@ -1,7 +1,14 @@
 #!/usr/bin/env bash
-# 외부 GPU 서버 최초 1회 셋업
-# 가정: NVIDIA GPU + CUDA 11.8 (다른 CUDA면 paddle wheel URL 수정)
-# 실행 위치: ~/workspace/kr_lp_pgnet/
+# kr_lp_pgnet 전용 컨테이너 안에서 실행되는 환경 셋업 스크립트.
+#
+# 가정 (컨테이너 외부):
+#   - Ubuntu 24.04 베이스 컨테이너 `kr_lp_pgnet` (--gpus all, --restart unless-stopped)
+#   - 호스트 /home/cuuva/workspace ↔ 컨테이너 /workspace (bind mount)
+#   - 이 repo는 호스트 ~/workspace/kr_lp_pgnet/ 에 clone (= 컨테이너 /workspace/kr_lp_pgnet/)
+#   - paddle sm_120 wheel은 ~/workspace/wheels/ 안에 둠
+#
+# 실행 (호스트에서):
+#   docker exec kr_lp_pgnet bash /workspace/kr_lp_pgnet/scripts/setup_server.sh
 
 set -euo pipefail
 
@@ -10,37 +17,63 @@ KR_LP_DIR="$(cd "$SCRIPT_DIR/.." && pwd)"
 WORKSPACE="${WORKSPACE:-$(cd "$KR_LP_DIR/.." && pwd)}"
 PADDLEOCR_DIR="$WORKSPACE/PaddleOCR"
 PRETRAIN_DIR="$PADDLEOCR_DIR/pretrain_models"
+WHEEL_DIR="$WORKSPACE/wheels"
 
-echo "[1/4] PaddlePaddle GPU 설치 (CUDA 11.8 기준)"
-python3 -m pip install --upgrade pip
-python3 -m pip install paddlepaddle-gpu==2.6.1.post118 \
-    -f https://www.paddlepaddle.org.cn/whl/linux/mkl/avx/stable.html
+PY=python3.10
 
-echo "[2/4] PaddleOCR clone"
+echo "[1/6] python / paddle 확인"
+$PY --version
+if ! $PY -c 'import paddle; assert paddle.is_compiled_with_cuda()' 2>/dev/null; then
+    echo "  paddle 미설치 또는 CUDA 비호환. sm_120 wheel 재설치..."
+    WHL=$(ls "$WHEEL_DIR"/paddlepaddle_gpu-*-cp310-*linux_x86_64.whl 2>/dev/null | head -1)
+    if [ -z "$WHL" ]; then
+        echo "  wheel 파일을 $WHEEL_DIR/ 에 두고 다시 실행하세요." >&2
+        exit 1
+    fi
+    $PY -m pip install --quiet "$WHL"
+fi
+$PY -c 'import paddle; print("  paddle:", paddle.__version__, "cuda:", paddle.is_compiled_with_cuda())'
+
+echo "[2/6] PaddleOCR clone (release/2.7)"
 if [ ! -d "$PADDLEOCR_DIR" ]; then
     git clone --depth 1 -b release/2.7 https://github.com/PaddlePaddle/PaddleOCR.git "$PADDLEOCR_DIR"
 fi
 cd "$PADDLEOCR_DIR"
-python3 -m pip install -r requirements.txt
+echo "  PaddleOCR @$(git rev-parse --short HEAD)"
+
+echo "[3/6] PaddleOCR requirements 설치 (paddlepaddle 라인은 제외)"
+if [ -f requirements.txt ]; then
+    grep -viE '^(paddlepaddle|paddleocr)' requirements.txt > /tmp/kr_lp_req.txt || true
+    $PY -m pip install --quiet -r /tmp/kr_lp_req.txt || true
+fi
 
-echo "[3/4] PGNet Step1 pretrain weight 다운로드"
+echo "[4/6] PaddleOCR import smoke test"
+cd "$PADDLEOCR_DIR"
+$PY -c "
+import sys; sys.path.insert(0, '.')
+import ppocr
+from ppocr.modeling.architectures import build_model
+print('  ppocr import OK')
+" || { echo "  PaddleOCR import 실패 (paddle 3.3 호환성 이슈일 수 있음)"; exit 1; }
+
+echo "[5/6] PGNet step1 pretrain weight 다운로드"
 mkdir -p "$PRETRAIN_DIR"
 cd "$PRETRAIN_DIR"
-if [ ! -d "train_step1" ]; then
+if [ ! -d train_step1 ]; then
     wget -q https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/train_step1.tar
-    tar xf train_step1.tar
-    rm train_step1.tar
+    tar xf train_step1.tar && rm train_step1.tar
 fi
+ls train_step1/
 
-echo "[4/5] kr_lp_dict 및 config를 PaddleOCR 트리에 symlink"
+echo "[6/6] dict / config symlink → PaddleOCR 트리"
 ln -sf "$KR_LP_DIR/dict/kr_lp_dict.txt" "$PADDLEOCR_DIR/ppocr/utils/kr_lp_dict.txt"
 mkdir -p "$PADDLEOCR_DIR/configs/e2e"
 ln -sf "$KR_LP_DIR/configs/kr_lp_pgnet.yml" "$PADDLEOCR_DIR/configs/e2e/kr_lp_pgnet.yml"
-
-echo "[5/5] GPU/Paddle 동작 확인"
-python3 -c "import paddle; print('paddle:', paddle.__version__); print('CUDA available:', paddle.is_compiled_with_cuda()); print('GPU count:', paddle.device.cuda.device_count())"
+ls -l "$PADDLEOCR_DIR/ppocr/utils/kr_lp_dict.txt" "$PADDLEOCR_DIR/configs/e2e/kr_lp_pgnet.yml"
 
 echo
+echo "==========================="
 echo "셋업 완료. 다음 단계:"
-echo "  python data_gen/generate_synthetic.py ..."
-echo "  bash scripts/run_step1.sh"
+echo "  bash $KR_LP_DIR/data_gen/setup_assets.sh        # 합성 자산 다운로드"
+echo "  python data_gen/generate_synthetic.py ...       # 합성 데이터 생성 (다음 단계)"
+echo "==========================="