StarCoder

HuggingFace's open code LLM trained on The Stack dataset

Code & DevelopmentFree

StarCoder is a family of open-source large language models for code, developed by HuggingFace and ServiceNow through the BigCode project. Trained on 80+ programming languages from The Stack dataset, StarCoder2-15B matches or exceeds GPT-3.5 on code benchmarks while being fully open-weight and commercially usable. It powers code completion in tools like VS Code's HuggingFace extension and serves as a baseline for fine-tuning on proprietary codebases.

Key Features

• ✓80+ languages
• ✓Open weights
• ✓15B parameters
• ✓Commercial license
• ✓Fine-tunable
• ✓The Stack training data