Atlas

Models

← All models

Gemma 1.1 2B IT

Google DeepMind · 2024-04-05 · 2.5B parameters

Gemma 1.1 2B IT is Google's instruction-tuned 2B-class checkpoint from the Gemma 1.1 update. It fine-tunes the original Gemma 2B model for chat and instruction-following use while remaining a compact open-weight Gemma model.

Benchmark scores

BenchmarkScore
Arena Text — Business, Management & Financial Ops (No Style Control)981
Arena Text — Chinese (No Style Control)1011
Arena Text — Creative Writing (No Style Control)998
Arena Text — English (No Style Control)1050
Arena Text — Entertainment, Sports & Media (No Style Control)980
Arena Text — Exclude Ties (No Style Control)797
Arena Text — Exclude Ties (Style Controlled)926
Arena Text — Expert (No Style Control)970
Arena Text — German (No Style Control)944
Arena Text — Hard Prompts (No Style Control)1005
Arena Text — Hard Prompts English (No Style Control)1028
Arena Text — Instruction Following (No Style Control)992
Arena Text — Korean (No Style Control)899
Arena Text — Legal & Government (No Style Control)995
Arena Text — Life, Physical & Social Science (No Style Control)996
Arena Text — Longer Query (No Style Control)1003
Arena Text — Medicine & Healthcare (No Style Control)990
Arena Text — Multi-turn (No Style Control)959
Arena Text — Non-English (No Style Control)988
Arena Text — Non-English (Style Controlled)1086
Arena Text — Occupational Mathematical (No Style Control)1043
Arena Text — Russian (No Style Control)989
Arena Text — Software & IT Services (No Style Control)1033
Arena Text — Writing, Literature & Language (No Style Control)1006
BBH (Open LLM Leaderboard v2)31.8
Capability88.6
GPQA (Open LLM Leaderboard v2)26.9
IFEval (Open LLM Leaderboard v2)30.7
LMArena Text Arena Chinese Elo1140
LMArena Text Arena Coding Elo1171
Loading Atlas data…