From 252e9d1cae22b46857988eec2297e03bbae03c86 Mon Sep 17 00:00:00 2001 From: "David S. Batista" Date: Thu, 20 Jun 2024 12:01:56 +0200 Subject: [PATCH] feat: LLM-based Evaluators have `raise_on_Failure=False` by default (#16) * initial import * adding partial --- .../evaluation/harness/rag/evaluation_pipeline.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/haystack_experimental/evaluation/harness/rag/evaluation_pipeline.py b/haystack_experimental/evaluation/harness/rag/evaluation_pipeline.py index 08c295c8..581a3150 100644 --- a/haystack_experimental/evaluation/harness/rag/evaluation_pipeline.py +++ b/haystack_experimental/evaluation/harness/rag/evaluation_pipeline.py @@ -44,8 +44,8 @@ def default_rag_evaluation_pipeline( RAGEvaluationMetric.SEMANTIC_ANSWER_SIMILARITY: partial( SASEvaluator, model="sentence-transformers/all-MiniLM-L6-v2" ), - RAGEvaluationMetric.ANSWER_FAITHFULNESS: FaithfulnessEvaluator, - RAGEvaluationMetric.CONTEXT_RELEVANCE: ContextRelevanceEvaluator, + RAGEvaluationMetric.ANSWER_FAITHFULNESS: partial(FaithfulnessEvaluator, raise_on_failure=False), + RAGEvaluationMetric.CONTEXT_RELEVANCE: partial(ContextRelevanceEvaluator, raise_on_failure=False), } for metric in metrics: