Llama 3
Meta
partial detail 10 method fields published
Evaluation
FP8 quantization effect on response quality
Effect of FP8 quantization errors on Llama 3 405B responses; standard benchmarks found inadequate for reflecting FP8 effects (occasional corrupted responses when scaling factors not upper bounded); reward distribution used instead
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.