A package to evaluate factuality of long-form generation. Original implementation of our EMNLP 2023 paper "FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation"
@martiansideofthemoon here are the dependency changes i mentioned to the pyproject.toml. It might be possible to further loosen the requirements, tough supporting torch>2.0 and getting rid of the mandatory install of pysqlite-binary was enough for my workflow. Since python already comes with a sqlite3 module, there's no need to enforce the use of pysqlite-binary (which as I mentioned in Slack just failed to install for me due to a missing wheel on my platform). If someone does want it installed, they can do so with pip install factscore[pysqlite-binary]
@martiansideofthemoon here are the dependency changes i mentioned to the pyproject.toml. It might be possible to further loosen the requirements, tough supporting
torch>2.0and getting rid of the mandatory install ofpysqlite-binarywas enough for my workflow. Since python already comes with a sqlite3 module, there's no need to enforce the use ofpysqlite-binary(which as I mentioned in Slack just failed to install for me due to a missing wheel on my platform). If someone does want it installed, they can do so withpip install factscore[pysqlite-binary]