arkhangelos/gemma-2b-scoring
The arkhangelos/gemma-2b-scoring model is a 2.6 billion parameter, decoder-only, text-to-text large language model from Google's Gemma family, built using the same research as Gemini models. It is instruction-tuned and available in English, excelling at various text generation tasks including question answering, summarization, and reasoning. Its compact size and open weights make it suitable for deployment in resource-limited environments like laptops or desktops, democratizing access to advanced AI capabilities.
Loading preview...
Model Overview
arkhangelos/gemma-2b-scoring is a 2.6 billion parameter instruction-tuned model from Google's Gemma 2 family. These lightweight, decoder-only, text-to-text models are developed from the same research and technology as the Gemini models. They are designed for a variety of text generation tasks in English, including question answering, summarization, and reasoning.
Key Capabilities
- Efficient Deployment: Its relatively small size (2.6B parameters) allows for deployment in environments with limited resources, such as laptops, desktops, or personal cloud infrastructure.
- Versatile Text Generation: Proficient in tasks like generating creative text formats, powering chatbots, and summarizing documents.
- Responsible AI Focus: Developed with a strong emphasis on Responsible AI, incorporating rigorous CSAM and sensitive data filtering during training, and undergoing extensive ethics and safety evaluations.
- Optimized Performance: Trained on Google's TPUv5p hardware using JAX and ML Pathways, enabling faster and more efficient training.
Training Details
The 2B model was trained on 2 trillion tokens, comprising a diverse dataset of web documents, code, and mathematical text to enhance linguistic styles, programming patterns, and logical reasoning. Data preprocessing included rigorous filtering for CSAM and sensitive personal information.
Usage Considerations
While offering strong performance for its size, users should be aware of potential limitations related to training data biases, context complexity, language ambiguity, and factual accuracy. The model is intended for content creation, communication, research, and educational applications, with guidelines provided for responsible use.