-
Notifications
You must be signed in to change notification settings - Fork 1.3k
Add Seed-OSS-36B-Instruct model integration #1182
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.
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
base: main
Are you sure you want to change the base?
Conversation
|
Was about to make a similar PR until I saw this! Thanks for doing this |
|
|
||
| formatted_prompt += "For each function call, return the function call in the exact format:\n" | ||
| formatted_prompt += "[function_name(parameter1=value1, parameter2=value2)]\n\n" | ||
| formatted_prompt += "IMPORTANT:\n" |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Hi @michyli , where did this section come from? For example, the word IMPORTANT never shows up in the Jinja2 template you pasted in the doc string. The _format_prompt function logic seems very different than the Jinja2 template.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
So the Jinja template is Seed's full native format, and it uses XML-style tool calls and includes functionalities not needed for leaderboard evals, like the think budget. The leaderboard expects function calls in python-style format ([function_name(param=value)]), which is different than the native XML style tool calls Seed has. The IMPORTANT section is there to ensure the output format of the model matches what the leaderboard expects, and not default to its native XML format or generate placeholder function name, causing evals to fail. I've had previous runs without this section and the output format doesn't match the expected and fails the evals.
Adds support for ByteDance Seed-OSS-36B-Instruct model