Model crashes while getting Word Embeddings

[moayadeldin]

When I try to get word embeddings for a dataset using the approach mentioned in Arabic Embeddings Notebook on Colab for relatively big input (+100 sentences) then Google Colab completely crashes as it ran out of memory. I tried to use my own PC but it crashed as well.
Is there any way to solve this problem other than dividing my dataset into very small pieces?
TIA.

[WissamAntoun]

try adding a torch no grad context.

with torch.no_grad():
    outputs = model(**inputs)

Colab should be able to handle 100 sentences

[moayadeldin]

Thank You! I used the same approach to handle much more sentences (+3000) and unfortunately it crashed, is there a way allowing me to handle many thousands of sentences without having this ram crash?

[WissamAntoun]

did you move the embeddings to CPU or it's crashing the GPU RAM?

Also, try saving the results on disk

[moayadeldin]

It crashes while using either GPU or CPU of Google Colab, and kindly tell me if there is a way to save results on disk while the model is running instead of RAM as notebook crashes before model finishes getting its output.