Roman Fitzjalen

I would be great if you add context from dataset to train the model. Lets say I have question, answer and context. How can I train it? Looks like providing...

known_bits and kp

10

Hello! I cannot understand how to fill data.json with my parameters. First of all I can't understand how can I find known_type, known_bits and kp? And hash in data.json is...

Prediction

1

Ok but how can we predict it in real life? for future period

Sometimes resampling is unnecessary (for example, in an ensemble custom predictor where all models are trained with the same spacing and there's a custom initial resampler). In such cases, it's...

I'm training gallstone and kidney stone model with pretrained weights and QuaterLR (using 250 epochs), momentum 0.95 and got this sharp loss decrease. It's not first time I see this....

Use real latest checkpoint in training

2

Solution for issue: https://github.com/MIC-DKFZ/nnUNet/issues/2563 Instead of loading the latest checkpoint when the flag `--c` is handled, we will filter to get the most recent checkpoint. This could be the latest,...