Any interest in adding prolog for reasoning problems?

[ohpauleez]

There have been a few interesting papers of the past year that translated reasoning and logic problems into prolog to enhance the capacity of various llms. Some examples:

• Reliable Reasoning Beyond Natural Language - github
• Neuro-Symbolic Integration Brings Causal and Reliable Reasoning Proofs - github.

While optillm has Z3 for reasoning, what are your thoughts about expanding these capabilities with the techniques from the paper, using something simple like pytholog (as done in one of the papers) or using the Python bridge to SWI Prolog (as done in the other paper)?

[codelion]

This would be a good addition, the z3 reasoning implemented in optillm now also includes sympy. It will be interesting to add more solvers/formal tools that can help with reasoning. Usually, when adding a new plugin or approach I look for a particular benchmark or eval suite to work with and compare with the baselines implemented in optillm already.