In instructGPT, during the RM training process, different <prompt, response> pairs of a prompt are put together to calculate the loss. Is this also implemented in DeepSpeed-chat? #320

Open

Open

In instructGPT, during the RM training process, different <prompt, response> pairs of a prompt are put together to calculate the loss. Is this also implemented in DeepSpeed-chat?#320

Assignees

Labels

deespeed chatquestion

No description provided.

Metadata

Assignees

jeffra

Labels

deespeed chatquestion

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests