phi3:medium

Microsoft's 14B parameter Small Language Model (SLM): delivers state-of-the-art performance, surpassing models like Gemini 1.0 Pro, optimized for reasoning and instruction-following.

Phi-3 Medium is a 14-billion parameter, dense decoder-only Transformer from Microsoft: a powerful Small Language Model (SLM) built for commercial and research deployment. It leverages a rigorous training pipeline, including 4.8 trillion tokens of high-quality, reasoning-dense data, Supervised Fine-Tuning (SFT), and Direct Preference Optimization (DPO). Available in 4K and 128K context variants, the model demonstrates state-of-the-art performance among lightweight models. Its core strengths lie in complex reasoning, mathematics, and precise instruction adherence, with benchmark results (e.g., MMLU-Pro) approaching the Llama 3 70B level.

https://huggingface.co/microsoft/Phi-3-medium-4k-instruct