OpenAI Evals
StaleDescription
OpenAI's framework for evaluating LLMs and LLM systems, providing an open-source registry of benchmarks and tools for systematic model assessment.
Key Features
- Open-source registry of evals for testing different dimensions of LLM performance
- Custom eval creation using basic and model-graded templates without writing code
- Private evals support for evaluating LLM patterns in your workflow without exposing data
- Completion Function Protocol for advanced use cases like prompt chains and tool-using agents
- Snowflake integration for logging eval results to a database
- Configurable and runnable directly in the OpenAI Dashboard
Use Cases
💡 Evaluate and benchmark different LLM model versions for your use case
💡 Build custom evals to test specific LLM capabilities you care about
💡 Red-team and safety testing of LLM systems
💡 Track model performance improvements across iterations
💡 Contribute evaluation benchmarks to the OpenAI community
Categories
Quick Start
1. Install: pip install evals
2. Set API key: export OPENAI_API_KEY=your_key
3. Download evals: git lfs fetch --all && git lfs pull
4. Run: oaieval <eval_name> --api_base <api_base>
5. Or create custom evals: pip install -e . and follow docs/build-eval.md