Ensembling small open-source LLMs as relevance judges achieves human-correlation scores competitive with GPT-4-based judges on the LLMJudge benchmark.
In: 2019 Seventh International Symposium on Computing and Networking Workshops (CANDARW)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.IR 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
JudgeBlender: Ensembling Judgments for Automatic Relevance Assessment
Ensembling small open-source LLMs as relevance judges achieves human-correlation scores competitive with GPT-4-based judges on the LLMJudge benchmark.