Atlas

Models

← All models

Gemma 2 2B

Google · 2024-07-31 · 2.6B parameters

Gemma 2 2B is Google DeepMind's open-weight, text-to-text, decoder-only pre-trained checkpoint in the Gemma 2 family, released on July 31, 2024 alongside its instruction-tuned variant. Its 2,614,341,888-parameter checkpoint was trained on 2 trillion primarily English tokens with knowledge distillation from a larger teacher, and alternates 4,096-token local sliding-window attention with 8,192-token global attention.

Benchmark scores

BenchmarkScore
Capability81.9
Global MMLU Lite — Google Unspecified Language Cohort43.3
GPQA Diamond12.5
MGSM18.7
MMLU Pro22.2
Loading Atlas data…