Any plan/idea to support to inference using beam search?

#51
by GoZion - opened

Hello! I try to inference using beam search (num_beams > 1), but the model does not support that. Do you have any plan to support to inference using beam search? Or how should I modify the modeling code to support that? Looking forward to your reply!

Microsoft org

Hello @Gone! Yes, we definitely have plans to support beam search.

We just added support for training/fine-tuning with attention mask and GQA/MQA. The next step is to add support for Flash-Attention and finally generation with beam search!

Hello @Gone! Yes, we definitely have plans to support beam search.

We just added support for training/fine-tuning with attention mask and GQA/MQA. The next step is to add support for Flash-Attention and finally generation with beam search!

GoZion changed discussion status to closed
Your need to confirm your account before you can post a new comment.

Sign up or log in to comment