PaddlePaddle/PaddleOCR · error · ValueError

not find model file path {}

Error message

not find model file path {}

What it means

In utility.create_predictor, when --use_onnx is set, the model path must point to an actual .onnx file (not a directory of Paddle inference files). This ValueError fires when the given path does not exist on disk. Note the separate branch above it: if model_dir itself is None the code logs and sys.exit(0)s instead — so this error specifically means 'a path was given but nothing is there'.

Source

Thrown at tools/infer/utility.py:205

        model_dir = args.ser_model_dir
    elif mode == "re":
        model_dir = args.re_model_dir
    elif mode == "sr":
        model_dir = args.sr_model_dir
    elif mode == "layout":
        model_dir = args.layout_model_dir
    else:
        model_dir = args.e2e_model_dir

    if model_dir is None:
        logger.info("not find {} model file path {}".format(mode, model_dir))
        sys.exit(0)
    if args.use_onnx:
        import onnxruntime as ort

        model_file_path = model_dir
        if not os.path.exists(model_file_path):
            raise ValueError("not find model file path {}".format(model_file_path))

        sess_options = args.onnx_sess_options or None

        if args.onnx_providers and len(args.onnx_providers) > 0:
            sess = ort.InferenceSession(
                model_file_path,
                providers=args.onnx_providers,
                sess_options=sess_options,
            )
        elif args.use_gpu:
            sess = ort.InferenceSession(
                model_file_path,
                providers=[
                    (
                        "CUDAExecutionProvider",
                        {
                            "device_id": args.gpu_id,
                            "cudnn_conv_algo_search": "HEURISTIC",

View on GitHub (pinned to 2661c7c0ef)

Solutions

1. Verify the file exists and is an .onnx file: ls -la <path>
2. Convert a Paddle model first: paddle2onnx --model_dir inference/det --model_filename inference.pdmodel --params_filename inference.pdiparams --save_file det.onnx
3. Pass the full file path (dir/model.onnx), not the directory, when use_onnx is enabled

Example fix

# before
--use_onnx --det_model_dir=./inference/det_ppocrv3  # directory, not .onnx -> ValueError

# after
paddle2onnx --model_dir ./inference/det_ppocrv3 \
  --model_filename inference.pdmodel --params_filename inference.pdiparams \
  --save_file ./inference/det.onnx
--use_onnx --det_model_dir=./inference/det.onnx

import os
assert args.use_onnx is False or (args.det_model_dir and os.path.isfile(args.det_model_dir) and args.det_model_dir.endswith('.onnx')), \
    'with --use_onnx the model path must be an existing .onnx file'

try:
    utility.create_predictor(args, 'det', logger)
except ValueError as e:
    if 'not find model file path' in str(e):
        raise SystemExit(f'ONNX model missing: {args.det_model_dir} — convert with paddle2onnx first')
    raise

When it happens

Trigger: Running with --use_onnx and a model path that doesn't exist: typo, wrong extension, passing a directory when an .onnx file path is required, or not having converted/downloaded the ONNX model.

Common situations: Converting workflow to ONNX but forgetting to run paddle2onnx; passing the Paddle model directory (containing .pdmodel) unchanged to the ONNX path; downloading models to a different folder than referenced.

Related errors

• degenerate text crop (zero width/height)
• File not found: '{file_path}'
• {file_path}
• {} does not exist!
• {} does not exist!

AI-assisted analysis of PaddlePaddle/PaddleOCR@2661c7c0ef (2026-08-14). Data as JSON: /api/errors/a81b172ba847129c. Report an issue: GitHub.