Atlas

Models

← All models

Qwen1.5 7B Chat

Alibaba · 2024-02-04 · 7.7B parameters

Qwen1.5-7B-Chat is Alibaba's chat-tuned 7B model from the Qwen1.5 family, aligned for direct conversational and instruction-following use on top of the Qwen1.5-7B base checkpoint.

Benchmark scores

BenchmarkScore
AlpacaEval 2.0 (length-controlled win rate)14.7
AlpacaEval 2.0 (raw win rate)11.8
Arena Text — Business, Management & Financial Ops (No Style Control)1063
Arena Text — Chinese (No Style Control)1140
Arena Text — Creative Writing (No Style Control)1035
Arena Text — English (No Style Control)1105
Arena Text — Entertainment, Sports & Media (No Style Control)1030
Arena Text — Exclude Ties (No Style Control)908
Arena Text — Exclude Ties (Style Controlled)993
Arena Text — Expert (No Style Control)1054
Arena Text — Hard Prompts (No Style Control)1066
Arena Text — Hard Prompts English (No Style Control)1077
Arena Text — Instruction Following (No Style Control)1059
Arena Text — Legal & Government (No Style Control)1103
Arena Text — Life, Physical & Social Science (No Style Control)1077
Arena Text — Longer Query (No Style Control)1090
Arena Text — Medicine & Healthcare (No Style Control)1055
Arena Text — Multi-turn (No Style Control)1062
Arena Text — Non-English (No Style Control)1058
Arena Text — Non-English (Style Controlled)1122
Arena Text — Occupational Mathematical (No Style Control)1107
Arena Text — Russian (No Style Control)1005
Arena Text — Software & IT Services (No Style Control)1091
Arena Text — Writing, Literature & Language (No Style Control)1064
BBH (Open LLM Leaderboard v2)45.1
Capability101.6
EQ-Bench v254.4
EQ-Bench v2 + MAGI-Hard Combined48.0
GPQA (Open LLM Leaderboard v2)30.3
IFEval (Open LLM Leaderboard v2)43.7
Loading Atlas data…