Phi-1.5 - Lightweight Transformer Model for Text Generation and Coding

About Model

Phi-1.5 is a lightweight, open-source Transformer model developed by Microsoft, trained on the same dataset as Phi-1 but augmented with additional NLP synthetic data. It demonstrates near state-of-the-art performance in common sense reasoning, language understanding, and logic-based tasks among models under 10 billion parameters. Unlike larger models, Phi-1.5 has not undergone fine-tuning for instruction following or reinforcement learning from human feedback (RLHF). Instead, it serves as a base model for AI research, allowing the community to explore safety, bias mitigation, and model controllability. For safety, generic web-crawled data (e.g., Common Crawl) has been excluded, reducing direct exposure to harmful content. However, the model may still generate biased or harmful content, making it best suited for research and controlled applications.