Atlas

Models

← All models

Llama 3.1 Tülu 3 70B DPO

Allen Institute for AI (Ai2) · 2024-11-21 · 70.6B parameters

Llama 3.1 Tülu 3 70B DPO is Ai2's separately published DPO-stage checkpoint, before the final RLVR-trained Tülu 3 70B artifact.

Benchmark scores

BenchmarkScore
BBH (Open LLM Leaderboard v2)61.5
Capability124.3
GPQA (Open LLM Leaderboard v2)37.6
GPQA Diamond46.3
IFEval (Open LLM Leaderboard v2)82.8
MATH Level 542.7
MATH Level 5 (Open LLM Leaderboard v2)44.9
MMLU-Pro (Open LLM Leaderboard v2)46.3
MuSR (Open LLM Leaderboard v2)49.2
OTIS Mock AIME 2024-20254.4
Loading Atlas data…