IFEval is a new benchmark of 25 verifiable instruction types and ~500 prompts for objective, reproducible evaluation of LLMs' instruction-following abilities.
Towards better evaluation of instruction-following: A case-study in summarization
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2023 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Instruction-Following Evaluation for Large Language Models
IFEval is a new benchmark of 25 verifiable instruction types and ~500 prompts for objective, reproducible evaluation of LLMs' instruction-following abilities.