Adds humaneval+ data and mbpp+ eval from the evalplus project, to give us some more rigorous code evaluations.
Thanks to @Nanami18 (Muru) for writing most of this, I just tweaked it for open instruct. Currently a draft until I properly test this.
HumanEval plus is easy - just swap out the data source to the HumanEval+ released data, and it should work! It even works nicely with HumanEvalPack for better evaluation of instruction-tuned models. MBPP+ uses data from huggingface, so it has a slightly different setup.
Adds humaneval+ data and mbpp+ eval from the evalplus project, to give us some more rigorous code evaluations.
Thanks to @Nanami18 (Muru) for writing most of this, I just tweaked it for open instruct. Currently a draft until I properly test this.
HumanEval plus is easy - just swap out the data source to the HumanEval+ released data, and it should work! It even works nicely with HumanEvalPack for better evaluation of instruction-tuned models. MBPP+ uses data from huggingface, so it has a slightly different setup.