predict.py:25-28: Resolve issue with all-time negative...

The puzzle 30-025c8878 from #30 has to be resolved:

Lines 25 to 28 in 7d5d392

    
           @todo #30:90min Resolve issue with all-time negative prediction. 
        
            We should resolve issue with negative prediction that performed all the time 
        
            after model has been trained. Probably issue is inside train.py, 
        
            where we probably ignore or mismatch some crucial for training parameters.

The puzzle was created by @rultor on 23-Apr-24.

Estimate: 90 minutes, role: DEV.

If you have any technical questions, don't ask me, submit new tickets instead. The task will be "done" when the problem is fixed and the text of the puzzle is removed from the source code. Here is more about PDD and about me.

we probably need to look at the outputs in predict.py

optimizing learning rate: learning_rate=2e-5, and reduced batched sizes: per_device_train_batch_size=8 and per_device_eval_batch_size=32 showcased this result:

Raw logits: tensor([[ 3.3061, -3.8666]], grad_fn=<AddmmBackward0>)
Probabilities: tensor([[9.9923e-01, 7.6663e-04]], grad_fn=<SoftmaxBackward0>)
Predicted label: negative

we should check the result of the model inference on a list of examples.

@0pdd the puzzle #52 is still not solved.

@0pdd the only puzzle #52 is solved here.

	@todo #30:90min Resolve issue with all-time negative prediction.
	We should resolve issue with negative prediction that performed all the time
	after model has been trained. Probably issue is inside train.py,
	where we probably ignore or mismatch some crucial for training parameters.