Atlas

Models

← All models

Qwen2 7B Instruct

Alibaba · 2024-06-06 · 7.6B parameters

Qwen2-7B-Instruct is Alibaba's dense 7B-class instruction-tuned Qwen2 checkpoint, post-trained for chat and instruction following with multilingual, coding, mathematics, and reasoning capabilities. It was trained with 32,768-token sequences, while Qwen documents extrapolated operation up to 131,072 tokens with YaRN.

Benchmark scores

BenchmarkScore
BBH (Open LLM Leaderboard v2)55.4
BigCodeBench Complete42.1
BigCodeBench Instruct29.1
BigCodeBench-Hard Complete13.5
BigCodeBench-Hard Instruct10.1
Capability109.8
GPQA (Open LLM Leaderboard v2)29.8
IFEval (Open LLM Leaderboard v2)56.8
MATH Level 5 (Open LLM Leaderboard v2)27.6
MMLU-Pro (Open LLM Leaderboard v2)38.5
MuSR (Open LLM Leaderboard v2)39.3
ZeroEval34.3
ZeroEval CRUX37.9
ZeroEval GSM8K80.1
ZeroEval MATH Level 523.9
ZeroEval MMLU-Redux66.9
ZeroEval ZebraLogic8.4
Loading Atlas data…