unsloth/Phi-3.5-mini-instruct
The unsloth/Phi-3.5-mini-instruct is a 3.8 billion parameter instruction-tuned decoder-only Transformer model developed by Microsoft AI and the Phi team. It supports a 128K token context length and is optimized for strong reasoning capabilities, particularly in code, math, and logic. This model excels in memory/compute constrained environments and latency-bound scenarios, offering competitive multilingual performance.
Loading preview...
unsloth/Phi-3.5-mini-instruct: Optimized for Reasoning and Multilingual Tasks
This model is a 3.8 billion parameter instruction-tuned variant of the Phi-3.5 family, developed by Microsoft AI. It leverages synthetic data and filtered public datasets, with a strong focus on high-quality, reasoning-dense information. The model supports an extensive 128K token context length, making it suitable for long document summarization and complex QA tasks.
Key Capabilities
- Strong Reasoning: Excels in code, math, and logic, achieving high scores on benchmarks like GSM8K and MATH.
- Multilingual Performance: Demonstrates competitive performance across various languages on benchmarks such as Multilingual MMLU and MGSM, despite its compact size.
- Long Context Understanding: Capable of handling 128K token contexts, outperforming some larger models in tasks like GovReport and QMSum.
- Efficient Fine-tuning: Unsloth provides tools for 2x faster fine-tuning with 50% less memory usage.
Good For
- Applications requiring strong reasoning in resource-constrained environments.
- Use cases demanding long context processing, such as document analysis and summarization.
- Multilingual applications where a compact yet capable model is needed.
- As a building block for generative AI features, especially when augmented with RAG for factual knowledge.