Model Overview
The microsoft/LLaMA-2-13b-GTL-Delta is a 13 billion parameter model from Microsoft's Industrial Foundation Models (IFMs) family, built upon the LLaMA-2 architecture. It has been extensively fine-tuned on over 380 diverse tabular datasets, making it specialized for industrial applications. The model processes language-formatted tabular data as input prompts and generates predictive answers for both classification and regression tasks.
Key Capabilities
- Tabular Data Processing: Optimized for analyzing and making predictions from structured tabular data.
- Industrial Applications: Designed for use across various industrial sectors, including commerce, healthcare, energy, and sustainability.
- Classification and Regression: Supports both classification and regression tasks based on the input prompt's instructions.
- In-context Learning: Leverages in-context learning to interpret and respond to prompts containing context data, task instructions, or direct questions.
Intended Uses
This model is best suited for scenarios requiring accurate predictions from diverse tabular datasets in industrial contexts. Users should format prompts to include feature descriptions, label descriptions, and the data itself, often in a table format, to guide the model effectively. The model's tokenizer supports a vocabulary size of up to 32,000 tokens, consistent with the base LLaMA-2 model.
Responsible AI Considerations
Like other large language models, LLaMA-2-13b-GTL-Delta may exhibit biases due to its training data, potentially leading to biased predictions in specific industrial scenarios. Users are advised to verify predictions with other sources or domain experts, especially in high-risk applications. Developers should implement responsible AI practices, including assessing suitability for consequential impacts, mitigating misinformation risks, and preventing harmful content generation or misuse.