Phi-3-mini's performance on reasoning skills is evaluated against phi-2, Mistral-7b-v0.1, Mixtral-8x7b, Gemma 7B, Llama-3-instruct8b, and GPT-3.5. The evaluation utilized a consistent pipeline across all models, ensuring comparability. Results reflect both common sense and logical reasoning capacities, achieved through few-shot prompting at a fixed temperature setting. The specific number of examples per benchmark is documented. These results may vary from others published due to differences in evaluation methods and optimization processes being absent for phi-3 models.
The results for phi-3-mini on standard open-source benchmarks measure the model's reasoning ability, comparing it to phi-2 and several other notable models.
All reported numbers for phi-3-mini were produced using the same pipeline to ensure comparability across model evaluations.
Collection
[
|
...
]