PaddlePaddle/PaddleOCR · error · ValueError

Input name {repr(name)} not found in `dynamic_shapes`

Error message

Input name {repr(name)} not found in `dynamic_shapes`

What it means

The inverse of the extra-key check: during TRT conversion, every actual model input must have an entry in dynamic_shapes, because TensorRT needs min/opt/max ranges for each input to build the engine. This ValueError fires when a model input has no corresponding key — e.g. a shape spec written for an older export with fewer/differently named inputs.

Source

Thrown at tools/infer/utility.py:479

        config = inference.Config(str(model_file), str(params_file))
        config.enable_use_gpu(100, device_id)
        # NOTE: Disable oneDNN to circumvent a bug in Paddle Inference
        config.disable_mkldnn()
        config.disable_glog_info()
        return inference.create_predictor(config)

    dynamic_shape_input_data = dynamic_shape_input_data or {}

    predictor = _get_predictor(pp_model_file, pp_params_file)
    input_names = predictor.get_input_names()
    for name in dynamic_shapes:
        if name not in input_names:
            raise ValueError(
                f"Invalid input name {repr(name)} found in `dynamic_shapes`"
            )
    for name in input_names:
        if name not in dynamic_shapes:
            raise ValueError(f"Input name {repr(name)} not found in `dynamic_shapes`")
    for name in dynamic_shape_input_data:
        if name not in input_names:
            raise ValueError(
                f"Invalid input name {repr(name)} found in `dynamic_shape_input_data`"
            )

    trt_inputs = []
    for name, candidate_shapes in dynamic_shapes.items():
        # XXX: Currently we have no way to get the data type of the tensor
        # without creating an input handle.
        handle = predictor.get_input_handle(name)
        dtype = _pd_dtype_to_np_dtype(handle.type())
        min_shape, opt_shape, max_shape = candidate_shapes
        if name in dynamic_shape_input_data:
            min_arr = np.array(dynamic_shape_input_data[name][0], dtype=dtype).reshape(
                min_shape
            )
            opt_arr = np.array(dynamic_shape_input_data[name][1], dtype=dtype).reshape(

View on GitHub (pinned to 2661c7c0ef)

Solutions

1. List the model's input names and ensure dynamic_shapes has a min/opt/max triple for each one
2. Update inference.yml trt_dynamic_shapes to cover all inputs with correct names
3. Re-download the model package that matches the inference code version

Example fix

# before — model inputs are ['images'] but yml has:
trt_dynamic_shapes: {x: [...]}  # -> Input name 'images' not found in `dynamic_shapes`

# after
trt_dynamic_shapes:
  images: [[1,3,48,320],[1,3,48,640],[1,3,48,1280]]

input_names = set(predictor.get_input_names())
missing = input_names - set(dynamic_shapes)
assert not missing, f'dynamic_shapes missing inputs: {missing} — every model input needs [min, opt, max]'

try:
    _convert_trt(dynamic_shapes, model_file, params_file, ...)
except ValueError as e:
    if 'not found in `dynamic_shapes`' in str(e):
        raise SystemExit('add min/opt/max shapes for every input reported by get_input_names()')
    raise

When it happens

Trigger: dynamic_shapes covering only 'x' while get_input_names() returns ['x', 'conv_weight'] or ['images']; partial hand-editing of the shape config; models with multiple inputs (rare for OCR) configured for one.

Common situations: Model re-exported with an extra input; yml from a single-input model reused with a multi-input one; typos in one of several input names so only some resolve.

Related errors

• Invalid input name {repr(name)} found in `dynamic_shapes`
• Invalid input name {repr(name)} found in `dynamic_shape_inpu
• Configuration Error: 'trt_dynamic_shapes' must be defined in
• Unsupported data type: {pd_dtype}

AI-assisted analysis of PaddlePaddle/PaddleOCR@2661c7c0ef (2026-08-14). Data as JSON: /api/errors/b13794d834952854. Report an issue: GitHub.