Model Overview
The microsoft/LLaMA-2-7b-GTL-Delta is a 7 billion parameter Industrial Foundation Model (IFM) developed by Microsoft, built upon the LLaMA-2 architecture. This model has been extensively fine-tuned on over 380 diverse tabular datasets, making it highly specialized for industrial applications. It is part of the IFMs family, which also includes a 13B version.
Key Capabilities
- Tabular Data Processing: Designed to process and analyze various tabular datasets from sectors such as commerce, healthcare, energy, and sustainability.
- Classification and Regression: Supports both classification and regression tasks by interpreting language-formatted input prompts containing context data, task instructions, or direct questions.
- In-Context Learning: Optimized for in-context learning, allowing it to make predictions based on provided examples within the prompt.
- LLaMA-2 Base: Inherits the LLaMA-2 tokenizer with a vocabulary size of up to 32,000 tokens and a 4096 token context length.
Intended Uses
This model is primarily intended for accurate prediction in classification and regression tasks using diverse tabular data. It is particularly well-suited for scenarios where structured data needs to be analyzed and predictions generated based on patterns and relationships within that data. Users should format prompts according to the specified structure, including feature and label descriptions, and data tables, to leverage its in-context learning capabilities effectively.
Responsible AI Considerations
As with other language models, LLaMA-2-7b-GTL-Delta may exhibit biases (data and algorithmic), lead to misinterpretations, or inherit vulnerabilities from its base model. Microsoft advises users to be aware of these risks, verify predictions with domain experts, and apply responsible AI best practices, especially in high-risk scenarios or those with consequential impacts.