Skip to content

Problem with huggingface imports in kaggle tpu #5625

@level14taken

Description

@level14taken

❓ Questions and Help

Hello everyone,

When working in kaggle with xla and huggingface in latest environment, even importing transformers in a different order(which library is imported first or last) or importing some models is causing errors,How do i tackle this?

kaggle-notebook to reproduce errors.

Additional info:-
This happens when import of libraries in not done in a certain way and it leads to error. This isn't in the notebook so posted it here.

2023-09-20 03:44:44.786788: F tensorflow/compiler/xla/xla_client/pjrt_computation_client.cc:81] Non-OK-status: pjrt::LoadPjrtPlugin( "tpu", sys_util::GetEnvString(env::kEnvTpuLibraryPath, "libtpu.so")) status: INVALID_ARGUMENT: Unexpected PJRT_Api size: expected 496, got 512. Check installed software versions.

@will-cromar @JackCaoG can you please help?

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