diff --git a/evals/registry/data/dutch-lexicon/samples.jsonl b/evals/registry/data/dutch-lexicon/samples.jsonl new file mode 100644 index 0000000000..67dab1837e --- /dev/null +++ b/evals/registry/data/dutch-lexicon/samples.jsonl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2990d7a68b286b6c48fdeab36f4951770cae0286f5b2f4b5e14556e846fda8eb +size 39678 diff --git a/evals/registry/evals/dutch-lexicon.yaml b/evals/registry/evals/dutch-lexicon.yaml new file mode 100644 index 0000000000..44bbd64392 --- /dev/null +++ b/evals/registry/evals/dutch-lexicon.yaml @@ -0,0 +1,9 @@ +dutch-lexicon: + id: dutch-lexicon.dev.v0 + description: Test the model's ability to distinguish between existing and often misspelled and hallucinated Dutch words. + metrics: [accuracy] + +dutch-lexicon.dev.v0: + class: evals.elsuite.basic.match:Match + args: + samples_jsonl: dutch-lexicon/samples.jsonl