Phi-3 Mini-Instruct is a 3.8 billion-parameter small language model (SLM) developed by Microsoft. It is part of the Phi-3 model family and is built on a dense decoder-only Transformer architecture. The model was trained on a dataset containing 3.3 trillion tokens, which includes a mix of high-quality synthetic data and filtered publicly available website content chosen for its reasoning-dense properties.
The model is instruction-tuned using Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO) to ensure it can follow complex instructions and maintain safety alignment. Despite its small parameter count, it is designed to achieve performance levels in reasoning, coding, and mathematics that are competitive with much larger models.
Phi-3 Mini-Instruct supports two primary context window variants: a standard 4K token version and an extended 128K token version. Its compact size and architecture make it suitable for efficient execution on resource-constrained environments, including mobile hardware and edge devices, while maintaining low latency for generative tasks.