Based on the reasoning approach in the STaR paper, @anacarsi and I found this python package which implements a tree-of-though algorithm.
We couldn't get it to work out of the box and got errors like AttributeError: module 'openai' has no attribute 'error' with openai==1.3.8 and RemoteDisconnected with openai==0.28. Also, downgrading would be incompatible with our current version of openai used in the LLMcoder.
Using a tree of thoughts approach could boost the reasoning capability and help solve the errors that occur in the completions.
@ArturHD warned us about being too dependent on 3rd party packages. It may also be feasable to implement a simple interface for tree of thoughts ourselves.
Based on the reasoning approach in the STaR paper, @anacarsi and I found this python package which implements a tree-of-though algorithm.
We couldn't get it to work out of the box and got errors like
AttributeError: module 'openai' has no attribute 'error'
withopenai==1.3.8
andRemoteDisconnected
withopenai==0.28
. Also, downgrading would be incompatible with our current version of openai used in the LLMcoder.Using a tree of thoughts approach could boost the reasoning capability and help solve the errors that occur in the completions.