a simple hack would be to save the model weights in every experiment if the code is not reproducible so that you can reproduce your inference score during evaluation. I am not really sure why it is not reproducible though , I will look into it

Wouldn't that affect me in the case of a review. That was one reason I avoided Fastai. I tried so many times. If you run your cell multiple times, do you still the get the same results?

The essence of saving the finetuned weights it to reproduce your score without retraining again since you already have the weights(There is no like time limit in this competition, therefore there is no need to retrain your training notebook to make sure it fits the time limit like in other competitions) . Infact it makes the review easier since they won't need to re-run your solution from the training phase, they will just do inference with the saved weights. I haven't tried rerunning either of my notebooks so I haven't tested if I get the same results.

Alright, makes perfect sense. Thank you @Koleshjr. Now let the competition begin🤣🔥

goodluck!

Good Luck too bro!

i am sufering from missing id error i need helper on this

Check whether the IDs in your subs follow the sample submission file. Then try using index = False in the block of code. Eg:

sub.to_csv(f"{CFG.model_name}_all_folds.csv", index= False)

Okay let me try🤝🤝