Move the CUDA sync caused by the max operation to a higher level#1532

Open

javak87 wants to merge 7 commits intoecmwf:developfrom

javak87:cuda_async_flash_atten_jan_26

Commits on Dec 28, 2025

mv max claculation to block.py classes
Javad Kasravi
committed
mv max calculation to engines
Javad Kasravi
committed
mv max calculation to encoder & model
Javad Kasravi
committed
mv max calculation to model
Javad Kasravi
committed
max calculation of tokens_nbors_lens & tcs_lens
Javad Kasravi
committed
ruff changes
Javad Kasravi
committed

Commits on Dec 29, 2025

run ./scripts/actions.sh lint
Javad Kasravi
committed