lenicolas

Hello, After updating to OpenNMT3.5.1, CT2 4.3.1 and applying the PR, I cannot obtain a functional model that uses RoPE and even with basic position encoding, there are be some...

Actually, this is a flash_atten2 issue : I have (carefully) merged yout PR's modifications in a CT2 version prior to flash_attention2 implementation, reverted to onmt-py 3.4.1 that doesn't use it...

Yes, it's tested and running, I handcopied from another screen and got confused with a fork still in test, but edited everything in the end before pulling it to main....

Hello pierotofy, Anything new with the PR? There have been lots of updates in LT and argostranslate these few months, I do not know how this goes with Locomotive (still...

Hello, More updates .... - evaluating pivots with the pivot_from and pivot_to args in eval (pivoting models should be saved as "en_xx" and "xx_en" in run_dir) - using spacy as...

Some little editions to ckfilter... changing some wrong former editions.

Sorry, I gitted directly from the "ddup" feature branch yesterday then pushed without looking and had to bin the previous PR. Deduplication is 100% reliable, tested on en-fr as well,...

Hi everybody, I've made it work. For starters, the platform: - Dell workstation with RTX6000 =Ada= GPU (flash-attn supports ONLY Hopper/Ampere/Ada, latest 2.7.4 doesn't support Ada/Linux) - **Windows 11 Pro...