[NOT MEANT TO MERG!] GRPO reward func for coding dataset

[August-murr]

refer to #28

This is an example of how to use the OpenCoders dataset to create a reward function for the GRPOTrainer. The reward function parses the generated code, creates an evaluation script, and executes it using E2B to calculate its accuracy by counting the number of test cases that passed, as well as measuring the execution time.

It's not perfect since there are edge cases and potential failures that have not been addressed.

I haven't been able to test it with GRPO due to other issues. I would appreciate it if anyone could try this out and see if there are any other issues or if it works properly.

A similar approach was used to train R1, although it likely relied solely on a LeetCode dataset.

[kalogyu]

Thanks for sharing the approach. I understand that this implementation is still in the testing phase and has some potential edge cases. I'll be happy to help test it out and look for any issues.

refer to #28

This is an example of how to use the OpenCoders dataset to create a reward function for the GRPOTrainer. The reward function parses the generated code, creates an evaluation script, and executes it using E2B to calculate its accuracy by counting the number of test cases that passed, as well as measuring the execution time.

It's not perfect since there are edge cases and potential failures that have not been addressed.

I haven't been able to test it with GRPO due to other issues. I would appreciate it if anyone could try this out and see if there are any other issues or if it works properly.

A similar approach was used to train R1, although it likely relied solely on a LeetCode dataset.