Could you please let me know if there are any codes or evaluation frameworks available to assist with using this dataset
Could you please let me know if there are any codes or evaluation frameworks available to assist with using this dataset
friend what are you mean exactly?
how to evaluate?
friend what are you mean exactly?
What I mean is, are there any evaluation codes available here? Or, what is the prompt used for evaluating this dataset? It seems different from the structure of datasets like HumanEval or MBPP, so I'm not sure how to use this data to construct prompts to input into the model. Thanks a lot！
Hi there! Tom from Cohere here. Apologies that handling this has taken so long. First of all, I wanted to let you know that we have just released v2 which adds more tests for C++, Java, Javascript, Go and Rust.
We are working on supporting evaluation via the BigCodeBench evaluation harness for the Python split. We don't have an eval harness ready to release, but you can run evaluation by running the new test_file attribute in the dataset (which describes a complete unit testing file) with the code generation in the same folder under an appropriate name (e.g., for Python it is code.py and see the data for info on other languages). Multilingual code evaluation is hard to cover all edge cases so we are not ready to share our setup yet. We will update the dataset when we have something to share. Thanks for your understanding!