Uh oh!
There was an error while loading. Please reload this page.
Added TimeWarp and a training data saving class for collecting TimeWarp data - #330
Conversation
| elif task_name.startswith("weblinx"): | ||
| import weblinx_browsergym | ||
| elif task_name.startswith("timewarp"): | ||
| import browsergym.timewarp |
There was a problem hiding this comment.
It would be great if you can update the readme table to include timewarp.
| """ | ||
| GenericAgent with training data saving functionality. | ||
| This module extends GenericAgent to save training data (system prompt, user prompt, and agent output) |
There was a problem hiding this comment.
Thank you for adding a new agent to AgentLab. I would keep this contribution as a separate PR from the Timewarp benchmark. It looks like we can extract step-wise (sys_prompt, user_prompt, and agent_output) from the saved traces using ExpResult. Feel free to write a utility function using that if needed. Currently, this change modifies loop.py (beyond adding timewarp), which we would prefer to avoid if possible.
Thanks again @amanjaiswal73892. Reverted to the original loop.py and updated the readme. Please lmk if there's anything else that I need to do on my end. |
Uh oh!
There was an error while loading. Please reload this page.
Adding TimeWarp so that AgentLab can be used with BrowserGym
The training data saving class is needed to use TimeTraj, which does knowledge distillation on data generated by another agent