How to run Jobbert as a quick example

[chalagashvili]

Hi there, this is not an issue but a question. Sorry if you have some other channel for questions and I do not know. Basically what I am looking for is an example how I can run https://huggingface.co/jjzha/jobbert-base-cased quickly to just see how good the existing model at Huggingface is. I would provide a job description and to see the extracted skills. Is there a quick way to see this somehow?

I tried to train models myself at google colab with GPU enabled torch.cuda.is_available() outputting true . Installed requirements as provided in Readme and then ran:

bash scripts/run.individual.sh jobbert skills 1

Unfortunately I receive a torch gpu error below:

Training jobbert on skills
/root/.local/lib/python3.7/site-packages/torch/cuda/__init__.py:52: UserWarning: CUDA initialization: Unexpected error from cudaGetDeviceCount(). Did you run some cuda functions before calling NumCudaDevices() that might have already set an error? Error 803: system has unsupported display driver / cuda driver combination (Triggered internally at  /pytorch/c10/cuda/CUDAFunctions.cpp:100.)
  return torch._C._cuda_getDeviceCount() > 0

So the training is done with CPU, and it tells me that approximate training time is 6 hours+.. I can't wait that much (with my internet being unstable..)

Or maybe if you could provide a jobbert skills model that would work also.

[jjzha]

Hi,

Not sure what triggered the error, but could you try running the line in run.individual.sh as standalone and explicitly with the --device flag?

This would be in your case:

python3 machamp/train.py --dataset_config configs/Skills/skills.json --parameters_config configs/Skills/jobbert.1.json --name skill.jobbert.skills.1 --device 0

Let me know if this works.

[chalagashvili]

The script that you provided worked without errors and is training is now running with GPU. Thank you!