Skip to content

expand teacher prompt tensors by num_generations in _prepare_teacher_logprob_inputs#122

Open
zhuxiaoxuhit wants to merge 1 commit intowenet-e2e:mainfrom
zhuxiaoxuhit:fix/teacher-logprob-inputs-num-generations
Open

expand teacher prompt tensors by num_generations in _prepare_teacher_logprob_inputs#122
zhuxiaoxuhit wants to merge 1 commit intowenet-e2e:mainfrom
zhuxiaoxuhit:fix/teacher-logprob-inputs-num-generations

Conversation

@zhuxiaoxuhit
Copy link
Copy Markdown
Contributor

Fix batch dimension mismatch in the teacher path by expanding prompt tensors via repeat_interleave to support num_generations > 1.

@yuekaizhang
Copy link
Copy Markdown
Contributor

Thanks. However, for the kd_trainer, I think we will always set num_generation=1.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants