Hello, and thank you for releasing LiveCodeBench — it is a very helpful benchmark.
I have run my own model on LiveCodeBench v6 and generated an output file containing the model’s predictions (in JSON / JSONL format). I would like to ask whether this repository provides an official or recommended way to evaluate such a result file directly.
More specifically, is there an existing script or evaluation pipeline that takes a completed output file (e.g., predictions aligned with LCB v6 problems) as input and computes the benchmark metrics? If so, could you please point me to the relevant documentation or code? If not, I would appreciate any guidance on the intended evaluation workflow for externally generated results.
Thank you very much for your time and for maintaining this benchmark.
Hello, and thank you for releasing LiveCodeBench — it is a very helpful benchmark.
I have run my own model on LiveCodeBench v6 and generated an output file containing the model’s predictions (in JSON / JSONL format). I would like to ask whether this repository provides an official or recommended way to evaluate such a result file directly.
More specifically, is there an existing script or evaluation pipeline that takes a completed output file (e.g., predictions aligned with LCB v6 problems) as input and computes the benchmark metrics? If so, could you please point me to the relevant documentation or code? If not, I would appreciate any guidance on the intended evaluation workflow for externally generated results.
Thank you very much for your time and for maintaining this benchmark.