Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Not a bug, just missing a space in README.md #857

Merged
merged 1 commit into from
Feb 1, 2024
Merged

Not a bug, just missing a space in README.md #857

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion applications/DeepSpeed-Chat/training/step2_reward_model_finetuning/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,7 @@ Finetuning the Reward Model (RM) is more or less similar to Step-1 Supervised F

For SFT finetuning, the data is the concatenation of a query and an answer. However, for RM finetuning, each batch of data consists of two query-answer pairs, i.e., the same query with a high-score answer and a low-score answer. This also leads to the second difference as describe below.

👉**The training objective difference**
👉 **The training objective difference**

For RW, the training objective is the pairwise ranking score, i.e., for the two query-answer pairs, RM is supposed to give a higher score to the better answer. There are multiple ways to achieve this. In our implementation, we use either the end token of the sequence or the first padding token as the aggregated score and compare them. Others may also use the average score for the entire answer as an alternative.

Expand Down
Loading