MLflow - Phoenix

MLflow includes built-in support for Phoenix evaluators through its third-party scorer interface. This allows Phoenix users to run their existing evaluation metrics within MLflow’s mlflow.genai.evaluate() pipeline alongside experiment tracking and model management.

Using Phoenix Evaluators in MLflow#

Phoenix evaluators such as Hallucination, QACorrectness, and Toxicity can be used directly as MLflow scorers:

from mlflow.genai.scorers.phoenix import Hallucination, QACorrectnessimport mlflowresults = mlflow.genai.evaluate(    data=eval_dataset,    scorers=[        Hallucination(model="openai:/gpt-4o"),        QACorrectness(model="openai:/gpt-4o"),    ],)

For details on available scorers and configuration, see the MLflow Phoenix integration docs.