Overview

RuterNorway/Llama-2-13b-chat-norwegian is a 13 billion parameter language model developed by Ruter AI Lab. It is a fine-tuned variant of Meta's Llama 2 13b Chat model, specifically adapted for the Norwegian language. The model was trained for one epoch on a combination of Norwegian Alpaca and 15,000 samples of machine-translated data from OpenOrca, along with a small subset of custom instructional data.

Key Capabilities

• Norwegian Language Proficiency: Tuned to understand and generate text in Norwegian.
• Chat Assistant: Intended for commercial and research use as an assistant-like chat model.
• Instruction Following: Supports both the Llama 2 Chat prompt format and the Alpaca prompt format.

Intended Use Cases

• Summarization: Excels at summarizing Norwegian text.
• Question Answering: Performs well on question-answering tasks in Norwegian.
• Chat Applications: Suitable for conversational AI in Norwegian.

Limitations

• The model is an LLM, not a knowledge model; it does not inherently possess more information about Norway than its base model.
• Performance is generally better on summarization, Q&A, and chat tasks than on tasks requiring deep domain-specific knowledge or free-form answers.
• Training data includes machine-translated content, which may contain grammatical errors.
• Optimal results often require prompt tuning for specific applications.