Fix template when add_generation_prompt=true
#14
by
matteogeniaccio
- opened
The assistant prefix in the template is missing a newline,
this causes the assistant response to always begin with a newline, breaking multiturn conversations in derived projects