Benchmark
TREx
A subset of the LAMA dataset, constructed by automatically extracting factual triples from Wikipedia. It is used to evaluate the factual knowledge of language models.
Timeline (last 90 days)
Nov 10
The research team constructed controlled factual contrast samples from the LAMA/TREx dataset, and used the retrieval model's own token prediction head to compute prediction scores for the gold answer object and the alternative answer object.
Unverified50%