ultralytics/yolov5 · error · RuntimeError

failed to load ONNX file: {onnx}

Error message

failed to load ONNX file: {onnx}

What it means

export.py's TensorRT builder raises RuntimeError when trt.OnnxParser.parse_from_file() returns False, meaning the TensorRT parser could not digest the ONNX file. This happens inside export_engine after the network is created, so TensorRT itself is installed and working; the ONNX graph is the problem (corrupt file, unsupported op, or opset/IR version too new for the installed TensorRT).

Source

Thrown at export.py:666

    if verbose:
        logger.min_severity = trt.Logger.Severity.VERBOSE

    builder = trt.Builder(logger)
    config = builder.create_builder_config()
    if is_trt10:
        config.set_memory_pool_limit(trt.MemoryPoolType.WORKSPACE, workspace << 30)
    else:  # TensorRT versions 7, 8
        config.max_workspace_size = workspace * 1 << 30
    if cache:  # enable timing cache
        Path(cache).parent.mkdir(parents=True, exist_ok=True)
        buf = Path(cache).read_bytes() if Path(cache).exists() else b""
        timing_cache = config.create_timing_cache(buf)
        config.set_timing_cache(timing_cache, ignore_mismatch=True)
    flag = 1 << int(trt.NetworkDefinitionCreationFlag.EXPLICIT_BATCH)
    network = builder.create_network(flag)
    parser = trt.OnnxParser(network, logger)
    if not parser.parse_from_file(str(onnx)):
        raise RuntimeError(f"failed to load ONNX file: {onnx}")

    inputs = [network.get_input(i) for i in range(network.num_inputs)]
    outputs = [network.get_output(i) for i in range(network.num_outputs)]
    for inp in inputs:
        LOGGER.info(f'{prefix} input "{inp.name}" with shape{inp.shape} {inp.dtype}')
    for out in outputs:
        LOGGER.info(f'{prefix} output "{out.name}" with shape{out.shape} {out.dtype}')

    if dynamic:
        if im.shape[0] <= 1:
            LOGGER.warning(f"{prefix} --dynamic model requires maximum --batch-size argument")
        profile = builder.create_optimization_profile()
        for inp in inputs:
            profile.set_shape(inp.name, (1, *im.shape[1:]), (max(1, im.shape[0] // 2), *im.shape[1:]), im.shape)
        config.add_optimization_profile(profile)

    LOGGER.info(f"{prefix} building FP{16 if builder.platform_has_fast_fp16 and half else 32} engine as {f}")
    if builder.platform_has_fast_fp16 and half:

View on GitHub (pinned to 20d1d78a08)

Solutions

  1. Regenerate the ONNX from scratch in the same run so versions match: python export.py --weights yolov5s.pt --include onnx engine.
  2. Validate the file first: python -c "import onnx; m=onnx.load('yolov5s.onnx'); onnx.checker.check_model(m)".
  3. Downgrade the ONNX opset to one your TensorRT supports (export.py --opset 12) or upgrade TensorRT.
  4. If the file is corrupt, delete the .onnx and re-export; never hand-copy partial files.

Example fix

# before
python export.py --weights yolov5s.pt --include engine  # reuses stale/corrupt yolov5s.onnx

# after
rm yolov5s.onnx && python export.py --weights yolov5s.pt --include onnx engine --opset 12
Defensive patterns

Strategy: validation

Validate before calling

import onnx

def onnx_is_parsable(path: str) -> bool:
    model = onnx.load(path)  # raises if corrupt
    onnx.checker.check_model(model)
    return True

Try / catch

try:
    export_engine(...)  # or run export.py --include engine
except RuntimeError as e:
    if 'failed to load ONNX' in str(e):
        onnx.checker.check_model(onnx.load(onnx_path))  # diagnose
        re_export_onnx()  # regenerate and retry once

Prevention

When it happens

Trigger: Running export.py --include engine against an ONNX file produced by a newer onnx/onnxruntime than the installed TensorRT supports; a truncated/corrupt .onnx from an interrupted export; custom layers or ops not supported by trt.OnnxParser; pointing --weights at a hand-edited ONNX file.

Common situations: Mixed toolchain versions (torch 2.x + onnx opset 17 with TensorRT 7.x); reusing an old .onnx after upgrading TensorRT; CI runners where the ONNX export step partially failed but left a file; concatenating exports across machines.

Related errors

AI-assisted analysis of ultralytics/yolov5@20d1d78a08 (2026-08-15). Data as JSON: /api/errors/f12092af1205a199. Report an issue: GitHub.