Llama 3
Meta
partial detail 8 method fields published
Evaluation
Instruction following (IFEval)
IFEval: approximately 500 'verifiable instructions' with heuristic verification; average of prompt-level and instruction-level accuracy under strict and loose constraints
How it was run
Open a release to see the setup, scoring, and other details that the lab published. Only fields the lab actually disclosed are shown.