Gemma 3 (12B) Overview

This model is a 12 billion parameter variant of the Gemma 3 family, developed by Google DeepMind. It is a multimodal model capable of processing both text and image inputs to generate text outputs. Key features include a substantial 128K token context window and support for over 140 languages, making it versatile for global applications.

Key Capabilities

• Multimodal Processing: Accepts text and images (normalized to 896x896 resolution) as input.
• Extensive Context: Utilizes a 128K token context window for comprehensive understanding.
• Multilingual Support: Trained on data in over 140 languages.
• Diverse Task Performance: Excels in text generation, image understanding, question answering, summarization, and reasoning.
• Efficient Deployment: Designed for deployment in environments with limited resources, such as laptops or cloud infrastructure.

Benchmark Highlights (Gemma 3 PT 12B)

• Reasoning: Achieves 84.2 on HellaSwag (10-shot) and 72.6 on BIG-Bench Hard (few-shot).
• STEM & Code: Scores 74.5 on MMLU (5-shot) and 71.0 on GSM8K (8-shot).
• Multilingual: Reaches 64.3 on MGSM and 69.4 on Global-MMLU-Lite.
• Multimodal: Attains 82.3 on DocVQA (val) and 71.2 on VQAv2.

Good For

• Content Creation: Generating creative text formats, marketing copy, and email drafts.
• Conversational AI: Powering chatbots and virtual assistants.
• Research & Education: Serving as a foundation for VLM/NLP research and language learning tools.
• Image Analysis: Extracting and interpreting visual data for text communications.