samrere / pytortto Goto Github PK

It would be interesting to compare the end-to-end performance against Pytorch if the data access time were also factored in, considering it is possibly to load data into the CuPy like this:
https://xarray.dev/blog/xarray-kvikio

Of course, it would depend a lot on how the data and the training job interact....

Hello,

I was wondering if it's possible to create a CRNN model for text recognition using PyTortto? I tried to use CTCLoss from pytorch but created issues.

If not, is it possible for me to copy pytorch CTC loss function and LSTM model and implement manually into PyTortto? Or is there a lot of adjustments that required?

Good day, I wanted to run your code in Colab, but I get an error when importing.
Tried:
import tortto as tt
Received error:

AttributeError: module 'cupy' has no attribute 'set_printoptions'

The error is coming from line 24 of __init__.py of tortto:
cp.set_printoptions(precision=4)

Regards.

Hello,

Thanks for creating a great project. Can you explain why the Numpy minimum version is 1.24? I removed the check and seemed to work correctly with no errors?

Can you also provide an example of loading one of the Resnet models under examples and how you would do inference on images?