SwinliQ-AIs/Phi-3-mini-4k-instruct
SwinliQ-AIs/Phi-3-mini-4k-instruct is a 3.8 billion parameter, lightweight, instruction-tuned causal language model developed by Microsoft. It is part of the Phi-3 family, optimized for strong reasoning, especially in math and logic, and supports a 4K token context length. This model is designed for commercial and research use in memory/compute constrained environments and latency-bound scenarios.
Loading preview...
Model Overview
SwinliQ-AIs/Phi-3-mini-4k-instruct is a 3.8 billion parameter, instruction-tuned language model from Microsoft's Phi-3 family. It is a lightweight, state-of-the-art model trained on a high-quality dataset focusing on reasoning-dense properties, including synthetic and filtered public web data. The model has undergone supervised fine-tuning (SFT) and direct preference optimization (DPO) for improved instruction following and safety. It supports a 4K token context length.
Key Capabilities & Performance
- Strong Reasoning: Excels in common sense, language understanding, math, code, and logical reasoning, demonstrating robust performance among models under 13 billion parameters.
- Instruction Following: Significant gains in instruction following and structured output, including explicit support for the
<|system|>tag and improved multi-turn conversation quality. - Benchmarks: Achieves competitive scores on various benchmarks, including MMLU (70.9), BigBench Hard CoT (73.5), GSM8K CoT (85.7), and HumanEval (57.3), often outperforming larger models in its class.
- Optimized for Efficiency: Designed for memory/compute constrained environments and latency-bound scenarios.
Intended Uses
This model is suitable for broad commercial and research applications, particularly where:
- Resource Constraints: Memory or compute resources are limited.
- Low Latency: Applications require quick response times.
- Strong Reasoning: Tasks demand robust mathematical and logical reasoning capabilities.
While strong in reasoning, its smaller size means it has less capacity for extensive world knowledge, which can be augmented with external search engines. The model is primarily trained on English text, and performance may vary for other languages. It is available with cross-platform support via ONNX Runtime for various devices and hardware configurations.