as slack discussion -
original code would produce an incomplete file if the process is stopped before the loop completes, which is handled by using cached_file now.
tested with
python -m capreolus.run rank.searcheval with benchmark.name=msmarcopsg searcher.name=msmarcopsgbm25
as slack discussion - original code would produce an incomplete file if the process is stopped before the loop completes, which is handled by using
cached_file
now.tested with
got