framework support via use_bidirectional_attention - Cheers from your friends at Baseten
#3
by michaelfeil - opened
Marking it as "use_bidirectional_attention", would be helpful. I also added it to nvidia and gemma3, and helps integration in popular frameworks. https://huggingface.co/nvidia/llama-embed-nemotron-8b/commit/1acaf42b890bafa464ef9a58d1c0db0dd26120d4
michaelfeil changed pull request title from Cheers from your friends at Baseten to framework support via use_bidirectional_attention - Cheers from your friends at Baseten
Great point! Thank you for pointing it out!
hongliu9903 changed pull request status to merged