Open Jaren1907 opened 2 months ago
According to the replication results from the community, it is evident that GPT-4 and other models can work within this framework. For specific details, please refer to the replication results in https://github.com/SidU/MathBlackBox.
The work in the paper seems very powerful and can make significant progress on the 8B open-source model. Then, when using this method on a more powerful model and taking advantage of better instruction-following ability and reasoning ability, can relatively breakthrough mathematical reasoning results be achieved?