Using HuggingFace Transformers pipeline?

[Dijkie85]

Hi! Trying to port a Python project into Julia. Julia's Transformers.jl library, however, doesn't seem to support HuggingFace pipelines, so I'm turning to PythonCall in order to use the Python package transformers. I can import it ok with transformers = pyimport("transformers"), but I get the following error when trying to actually call transformers.pipeline:

julia> pos = transformers.pipeline("token-classification", model="PlanTL-GOB-ES/roberta-large-bne-capitel-pos")
ERROR: Python: RuntimeError: Failed to import transformers.pipelines because of the following error (look up to see its traceback):
/home/dijkie/packages/julias/julia-1.8/bin/../lib/julia/libstdc++.so.6: version `GLIBCXX_3.4.30' not found (required by /home/dijkie/miniconda3/envs/ds/lib/python3.10/site-packages/scipy/optimize/_highs/_highs_wrapper.cpython-310-x86_64-linux-gnu.so)
Python stacktrace:
 [1] _get_module
   @ transformers.file_utils ~/miniconda3/envs/ds/lib/python3.10/site-packages/transformers/file_utils.py:2779
 [2] __getattr__
   @ transformers.file_utils ~/miniconda3/envs/ds/lib/python3.10/site-packages/transformers/file_utils.py:2767
Stacktrace:
 [1] pythrow()
   @ PythonCall ~/.julia/packages/PythonCall/eU0yr/src/err.jl:94
 [2] errcheck
   @ ~/.julia/packages/PythonCall/eU0yr/src/err.jl:10 [inlined]
 [3] pygetattr(x::Py, k::String)
   @ PythonCall ~/.julia/packages/PythonCall/eU0yr/src/abstract/object.jl:60
 [4] getproperty(x::Py, k::Symbol)
   @ PythonCall ~/.julia/packages/PythonCall/eU0yr/src/Py.jl:272
 [5] top-level scope
   @ REPL[7]:1

So, there seems to be a problem with GLIBCXX_3.4.30 which I cannot decipher, much less correct.
Any ideas?

[cjdoris]

I thought I fixed this a while ago. Can you try upgrading?

[Dijkie85]

I thought I fixed this a while ago. Can you try upgrading?

Thank you for your quick reply!
What should I upgrade, and how?
I'm using Julia v. 1.8 and Python v. 3.10. Just installed PythonCall with ] add PythonCall a couple of hours ago...

[cjdoris]

If you haven't configured anything, then PythonCall should automatically install a version of glibcxx compatible with Julia. Are you setting any env vars to do with PythonCall or CondaPkg?

[Dijkie85]

Thank you again for taking the time to reply. Yes, I'musing PythonCall ENV["JULIA_CONDAPKG_BACKEND"] = "Current" to use my usual NLP environment. I ended up copying my system's libstdc++.so.6 to the Julia folder which includes the most up-to-date version of GLIBCXX and it worked. Kinda hacky though...