Hi Anubhab,
Thank you for delivering this great feature! I just wanted to check—have you tested it with model saving, loading, and retraining workflows?

I've defined loss_arr as a class attribute that tracks the loss value, during the fitting stage. Just checked:

1. model.plot_loss_arr() does work when a saved model is loaded.
2. It doesn't work as intended after re-training. If the initial training is for 10 epochs, and then fit_transform is applied again for 2 more epochs, we want the loss function to be plotted for all 12 epochs, right?

I've changed the logic (see commit 71c2f6f) such that in successive rounds of re-training, loss_arr list is extended, and not overwritten. This should fix the issue.

In the following code, I've also attached the outputs in the comments to demonstrate that it works for saving-loading & re-training workflows:

from fastopic import FASTopic

model = FASTopic(num_topics=50, verbose=True,device='cpu')
print(model.loss_arr) # Prints -> []

topic_top_words, doc_topic_dist = model.fit_transform(docs,epochs=10)
print(len(model.loss_arr)) # Prints -> 10

topic_top_words2, doc_topic_dist2 = model.fit_transform(docs,epochs=3)
print(len(model.loss_arr)) # Prints -> 13

model.save("fastopic.zip")

loaded_model = FASTopic.from_pretrained("fastopic.zip")
print(len(loaded_model.loss_arr)) # Prints -> 13

topic_top_words3, doc_topic_dist3 = loaded_model.fit_transform(docs,epochs=2)
print(len(loaded_model.loss_arr)) # Prints -> 15