feat: 同步报销流程与工作台改动

server/scripts/bootstrap_paddleocr_gpu.sh

@@ -0,0 +1,36 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+OCR_VENV_DIR="${OCR_VENV_DIR:-${ROOT_DIR}/.venv-ocr312}"
+PYTHON_BIN="${PYTHON_BIN:-python3.12}"
+PADDLEPADDLE_GPU_VERSION="${PADDLEPADDLE_GPU_VERSION:-3.3.0}"
+PADDLEOCR_VERSION="${PADDLEOCR_VERSION:-3.6.0}"
+PADDLE_GPU_INDEX_URL="${PADDLE_GPU_INDEX_URL:-https://www.paddlepaddle.org.cn/packages/stable/cu126/}"
+
+if ! command -v "${PYTHON_BIN}" >/dev/null 2>&1; then
+  echo "python3.12 不存在，请先安装 Python 3.12。" >&2
+  exit 1
+fi
+
+apt-get update
+apt-get install -y --no-install-recommends libgl1 libglib2.0-0 poppler-utils
+
+rm -rf "${OCR_VENV_DIR}"
+"${PYTHON_BIN}" -m venv "${OCR_VENV_DIR}"
+"${OCR_VENV_DIR}/bin/pip" install --upgrade pip
+"${OCR_VENV_DIR}/bin/pip" install \
+  "paddlepaddle-gpu==${PADDLEPADDLE_GPU_VERSION}" \
+  -i "${PADDLE_GPU_INDEX_URL}"
+"${OCR_VENV_DIR}/bin/pip" install "paddleocr==${PADDLEOCR_VERSION}"
+
+"${OCR_VENV_DIR}/bin/python" - <<'PY'
+import paddle
+
+print("PaddlePaddle:", paddle.__version__)
+print("CUDA compiled:", paddle.is_compiled_with_cuda())
+print("CUDA device count:", paddle.device.cuda.device_count())
+paddle.utils.run_check()
+PY
+
+echo "PaddleOCR GPU runtime ${PADDLEOCR_VERSION} 已安装到 ${OCR_VENV_DIR}"

server/scripts/paddle_ocr_worker.py

@@ -21,6 +21,7 @@ def parse_args() -> argparse.Namespace:
     parser.add_argument("--lang", default="ch")
     parser.add_argument("--text-detection-model", default="PP-OCRv5_mobile_det")
     parser.add_argument("--text-recognition-model", default="PP-OCRv5_mobile_rec")
+    parser.add_argument("--device", default=os.environ.get("OCR_DEVICE", ""))
     parser.add_argument("--enable-mkldnn", action="store_true")
     return parser.parse_args()
 
@@ -100,16 +101,20 @@ def build_document(input_path: str, results: list[Any]) -> dict[str, Any]:
 
 def main() -> int:
     args = parse_args()
-    ocr = PaddleOCR(
-        text_detection_model_name=args.text_detection_model,
-        text_recognition_model_name=args.text_recognition_model,
-        use_doc_orientation_classify=False,
-        use_doc_unwarping=False,
-        use_textline_orientation=False,
-        lang=args.lang,
+    ocr_options = {
+        "text_detection_model_name": args.text_detection_model,
+        "text_recognition_model_name": args.text_recognition_model,
+        "use_doc_orientation_classify": False,
+        "use_doc_unwarping": False,
+        "use_textline_orientation": False,
+        "lang": args.lang,
         # PaddlePaddle 3.3.x CPU oneDNN can fail on PP-OCRv5 static inference.
-        enable_mkldnn=args.enable_mkldnn,
-    )
+        "enable_mkldnn": args.enable_mkldnn,
+    }
+    configured_device = str(args.device or "").strip()
+    if configured_device:
+        ocr_options["device"] = configured_device
+    ocr = PaddleOCR(**ocr_options)
 
     documents = []
     for input_path in args.inputs: