Datasets & Experiments

Via Langfuse Datasets you can create test sets and benchmarks to evaluate the performance of your LLM application.

• Continuous improvement: Create datasets from production edge cases to improve your application
• Pre-deployment testing: Benchmark new releases before deploying to production
• Structured testing: Run experiments on collections of inputs and expected outputs
• Flexible evaluation: Add custom evaluation metrics or use llm-as-a-judge
• Integrates well: Works with popular frameworks like LangChain and LlamaIndex

How to build a workflow around datasets

This is a high-level example workflow of using datasets to continuously improve an LLM application:

1. Create dataset items with inputs and expected outputs through:

   • Manual creation or import of test cases
   • Synthetic generation of questions/responses
   • Production app traces with issues that need attention

2. Make changes to your application that you want to test

3. Run your application (or parts of it) on all dataset items

4. Evaluate results:

   • Compare against baseline/expected outputs if available
   • Use custom evaluation metrics
   • Leverage LLM-based evaluation

5. Review aggregated results across the full dataset to:

   • Identify improvements
   • Catch regressions
   • Make data-driven decisions about releases

Process diagram:

Data model

• Dataset is a collection of DatasetItems
  • DatasetItem contains input, expected_output, and metadata
• DatasetRun is an experiment run on a Dataset, it is identified by a unique name
  • DatasetRunItem links a DatasetItem to a Trace created during an experiment
  • Evaluation metrics of a DatasetRun are based on Scores associated with the Traces linked to run

FAQ

• I have setup Langfuse, but I do not see any traces in the dashboard. How to solve this?

GitHub Discussions