Skip to content

fix: tokenization mismatch when training with different templates #2996

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

Merged
merrymercy merged 1 commit into from
Feb 1, 2024
Merged

fix: tokenization mismatch when training with different templates #2996

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
8 changes: 5 additions & 3 deletions fastchat/train/train_with_template.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -132,9 +132,9 @@ def get_prompt_separator(conv):

elif conv.sep_style == SeparatorStyle.CHATML:
if conv.sep2 is None:
user_turn_separator = conv.sep
user_turn_separator = conv.sep + "\n"
else:
user_turn_separator = conv.sep2
user_turn_separator = conv.sep2 + "\n"

assistant_turn_separator = conv.roles[1] + "\n"

Expand All @@ -155,7 +155,9 @@ def mask_targets(conversations, targets, tokenizer, conv):
user_turn_separator, assistant_turn_separator = get_prompt_separator(conv)
turns = conversation.split(user_turn_separator)
for i, turn in enumerate(turns):
if turn == "":
if (
i < len(turns) - 1 and turn == ""
): # Last turn is the user_turn_separator
break

if i != 0:
Expand Down