Atlas

Models

← All models

Qwen1.5 72B Chat

Alibaba · 2024-02-04 · 72.3B parameters

Qwen1.5-72B-Chat is Alibaba's chat-tuned 72B model from the Qwen1.5 family, aligned for direct conversational use on top of the Qwen1.5-72B base checkpoint.

Benchmark scores

BenchmarkScore
AlpacaEval 2.0 (length-controlled win rate)36.6
AlpacaEval 2.0 (raw win rate)26.5
Arena Text — Business, Management & Financial Ops (No Style Control)1117
Arena Text — Chinese (No Style Control)1185
Arena Text — Creative Writing (No Style Control)1138
Arena Text — English (No Style Control)1187
Arena Text — Entertainment, Sports & Media (No Style Control)1126
Arena Text — Exclude Ties (No Style Control)1033
Arena Text — Exclude Ties (Style Controlled)1127
Arena Text — Expert (No Style Control)1135
Arena Text — French (No Style Control)1158
Arena Text — French (Style Controlled)1235
Arena Text — German (No Style Control)1084
Arena Text — Hard Prompts (No Style Control)1148
Arena Text — Hard Prompts English (No Style Control)1164
Arena Text — Instruction Following (No Style Control)1142
Arena Text — Japanese (No Style Control)1062
Arena Text — Korean (No Style Control)1050
Arena Text — Legal & Government (No Style Control)1173
Arena Text — Life, Physical & Social Science (No Style Control)1158
Arena Text — Longer Query (No Style Control)1157
Arena Text — Medicine & Healthcare (No Style Control)1136
Arena Text — Multi-turn (No Style Control)1160
Arena Text — Non-English (No Style Control)1135
Arena Text — Non-English (Style Controlled)1204
Arena Text — Occupational Mathematical (No Style Control)1172
Arena Text — Russian (No Style Control)1103
Arena Text — Software & IT Services (No Style Control)1162
Arena Text — Spanish (No Style Control)1109
Arena Text — Writing, Literature & Language (No Style Control)1159
Loading Atlas data…