Model Overview
SillyTilly/google-gemma-2-27b-it is a 27 billion parameter instruction-tuned model from Google's Gemma family, derived from the same research and technology as the Gemini models. It is a text-to-text, decoder-only large language model available in English, with open weights for both pre-trained and instruction-tuned variants. The model is designed for a wide range of text generation tasks, including question answering, summarization, and reasoning, and its relatively compact size allows for deployment on devices with limited resources.
Key Capabilities
- Text Generation: Excels at generating creative text formats, code, marketing copy, and email drafts.
- Conversational AI: Suitable for chatbots, virtual assistants, and interactive applications.
- Text Summarization: Can produce concise summaries of documents and research papers.
- Research & Education: Serves as a foundation for NLP research, language learning tools, and knowledge exploration.
Training Details
The 27B model was trained on 13 trillion tokens, encompassing a diverse dataset including web documents, code, and mathematical texts. This broad training ensures exposure to various linguistic styles, programming patterns, and logical reasoning. Data preprocessing included rigorous CSAM and sensitive data filtering, along with quality and safety checks.
Performance Highlights
Evaluated against numerous benchmarks, the Gemma 2 27B instruction-tuned model demonstrates strong performance across various tasks. Notable scores include 75.2 on MMLU, 86.4 on HellaSwag, 51.8 on HumanEval (pass@1), and 74.0 on GSM8K (maj@1). The model also shows competitive safety benchmark results, including 8.84 on RealToxicity and 86.94 on BBQ Disambig.
Intended Usage
This model is intended for content creation, communication, research, and educational applications. Its design prioritizes responsible AI development, offering high performance compared to other similarly sized open models.