Atlas

Models

← All models

Qwen1.5 14B Chat

Alibaba · 2024-02-04 · 14.2B parameters

Qwen1.5-14B-Chat is Alibaba's chat-tuned 14B model from the Qwen1.5 family, aligned for conversational and instruction-following tasks while retaining the 32K-context Qwen1.5 architecture.

Benchmark scores

BenchmarkScore
AlpacaEval 2.0 (length-controlled win rate)23.9
AlpacaEval 2.0 (raw win rate)18.6
Arena Text — Business, Management & Financial Ops (No Style Control)1083
Arena Text — Chinese (No Style Control)1146
Arena Text — Creative Writing (No Style Control)1091
Arena Text — English (No Style Control)1148
Arena Text — Entertainment, Sports & Media (No Style Control)1069
Arena Text — Exclude Ties (No Style Control)976
Arena Text — Exclude Ties (Style Controlled)1065
Arena Text — Expert (No Style Control)1093
Arena Text — French (No Style Control)1115
Arena Text — French (Style Controlled)1191
Arena Text — German (No Style Control)1042
Arena Text — Hard Prompts (No Style Control)1113
Arena Text — Hard Prompts English (No Style Control)1125
Arena Text — Instruction Following (No Style Control)1102
Arena Text — Japanese (No Style Control)1020
Arena Text — Legal & Government (No Style Control)1119
Arena Text — Life, Physical & Social Science (No Style Control)1103
Arena Text — Longer Query (No Style Control)1113
Arena Text — Medicine & Healthcare (No Style Control)1104
Arena Text — Multi-turn (No Style Control)1110
Arena Text — Non-English (No Style Control)1094
Arena Text — Non-English (Style Controlled)1160
Arena Text — Occupational Mathematical (No Style Control)1136
Arena Text — Russian (No Style Control)1045
Arena Text — Software & IT Services (No Style Control)1141
Arena Text — Spanish (No Style Control)1084
Arena Text — Writing, Literature & Language (No Style Control)1112
BBH (Open LLM Leaderboard v2)52.3
Loading Atlas data…