Yash Sachdeva

Hello, I have thoroughly analyzed the issue regarding inconsistent NaN handling in TensorFlow's sorting and indexing operations across CPU and GPU. I have implemented a comprehensive fix that ensures consistent...

Thank you for the clarification, @mihaimaruseac! You're absolutely right — NaN comparisons are undefined by standard, and variation across CPU/GPU is expected due to how sorting ops are implemented. However,...

Thank you for the clarification, @mihaimaruseac. Totally understand and respect TensorFlow’s stance on prioritizing performance over consistency for edge cases like NaN sorting. To align with this direction, I’d like...

I'd like to propose adding Dice Loss and Dice + Cross Entropy (CE) Loss functions to Optax. These are widely used in segmentation tasks, particularly in medical imaging and increasingly...

Hi @carlosgmartin @vroulet @q-berthet, I'd love to contribute and can implement both suggested changes if you're okay with it. Let me know if I should go ahead and open a...

Hi @zevla, This sounds like a useful enhancement! I'd love to help implement a PDF download option for the online documentation. Could you please guide me to the relevant repo...

Thanks for raising this! I think both points make sense, especially: - Using uv alongside pip could simplify reproducibility and align with modern dependency practices. - A Makefile would be...