Skip to content

Linux installation does not export location of libflexmlrt.so in LD_LIBRARY_PATH #386

Description

@Swiftloke

You would expect the venv/bin/activate script to set this when initializing the library paths, but it fails to do so:

LD_LIBRARY_PATH="${SITE_PACKAGES}/flexml/flexml_extras/lib:${SITE_PACKAGES}/onnxruntime/capi:${SITE_PACKAGES}/voe/lib:${LD_LIBRARY_PATH}"

Oops, it's not in any of those:

~/ryzenai$ find|grep libflexmlrt.so
./venv/lib/python3.12/site-packages/flexmlrt/lib/libflexmlrt.so
./venv/deployment/lib/libflexmlrt.so
./venv/onnxruntime/lib/libflexmlrt.so

Which causes the OCR example to fail, unless you add that path yourself:

onnxruntime.capi.onnxruntime_pybind11_state.RuntimeException: [ONNXRuntimeError] : 6 : RUNTIME_EXCEPTION : Exception during initialization: ERROR: Failed to create Model for /home/xyz/RyzenAI-SW/CNN-examples/Nemotron-OCR-V2/./cache/modelcachekey_nvidia-nemotron-ocr-v2-recognizer-english_1x128x8x32_bf16/vaiml_par_0, Reason: ERROR: Cannot find libflexmlrt.so in your conda environment . Please check if flexmlrt package is installed properly.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions