Wav2Vec2 Fine Tuning

Hello!I had written this code as a part of a hackathon. The hackathon is now over so be rest assured that I am not outsourcing my work to anyone.

We were tasked to make a multilingual ASR model and hence I chose Wav2Vec2 XLSR 300m for it.

Now, I followed a tutorial notebook - link to do this.

I feel like I have done everything correctly but for some reason, the final output is just not coming. The final step - trainer.train() is going under a infinite loop which is maxing out the CPU and not even using 1% of the GPU.

Here’s my notebook - Google Colab

To run it, just upload your kaggle.json to the session storage and run all the cells. The rest will be taken care of. When the execution reaches the final step, you will see that it is going for an infinite loop with 100% CPU and no utilization of the GPU.

If anybody can help me with this then it would really mean a lot to me. Since the hackathon is over, I would just like to learn from my mistakes and see what I have done wrong.

Thank you.