OpenAI evals - Phoenix

Using OpenAI with Phoenix Evals#

Info

Requires openai>=1.0.0

pip install "arize-phoenix-evals>=3" openai

Create an LLM instance with the OpenAI provider:

from phoenix.evals import LLMllm = LLM(provider="openai", model="gpt-4o")

The LLM wrapper reads your API key from the OPENAI_API_KEY environment variable, or you can pass it directly:

llm = LLM(provider="openai", model="gpt-4o", api_key="sk-...")

Using with evaluators#

from phoenix.evals import LLM, evaluate_dataframefrom phoenix.evals.metrics import FaithfulnessEvaluatorllm = LLM(provider="openai", model="gpt-4o")evaluator = FaithfulnessEvaluator(llm=llm)results_df = evaluate_dataframe(dataframe=df, evaluators=[evaluator])

Custom parameters#

Pass additional parameters to the OpenAI client:

llm = LLM(    provider="openai",    model="gpt-4o",    temperature=0.0,    sync_client_kwargs={"timeout": 60.0},    async_client_kwargs={"timeout": 120.0},)

Azure OpenAI#

Use the "azure" provider for Azure OpenAI deployments:

llm = LLM(    provider="azure",    model="gpt-4o",  # This is the deployment name    api_key="your-api-key",    api_version="2024-02-01",    base_url="https://your-resource.openai.azure.com/",)

Info

The model parameter is the deployment name in Azure. You can find it in the Azure OpenAI playground.

For full details on Azure OpenAI, check out the OpenAI Documentation.