Phi-1

Phi-1 is a small language model (SLM) developed by Microsoft, featuring 1.3 billion parameters. It is specifically designed and trained for Python code generation. Its training data includes a curated mix of Python code from public datasets like The Stack, question-and-answer content from StackOverflow, and synthetic Python textbooks generated by GPT-3.5. Despite its relatively small size compared to larger LLMs, Phi-1 demonstrates strong performance on coding benchmarks, notably exceeding 50% accuracy on HumanEval. The model is intended for research purposes and as a starting point for code generation tasks, not for direct use in production environments. It is available on the Hugging Face platform and can be used with the Transformers library.