> ## Documentation Index > Fetch the complete documentation index at: https://arize-ax.mintlify.dev/docs/llms.txt > Use this file to discover all available pages before exploring further. # LiteLLM > Use LiteLLM as the judge LLM proxy for evaluations with the arize-phoenix-evals library. The [`arize-phoenix-evals`](https://pypi.org/project/arize-phoenix-evals/) library uses an LLM-as-judge to grade model output — hallucinations, factuality, helpfulness, toxicity, custom rubrics. LiteLLM is the universal proxy provider in Phoenix Evals: pass `provider="litellm"` and a `model="/"` string to the `LLM(...)` wrapper to route the judge through any of [100+ LiteLLM-supported backends](https://docs.litellm.ai/docs/providers) — useful when no native Phoenix adapter exists (Mistral, Bedrock, Together, Groq, Ollama, etc.) or when you want one piece of eval code to switch backends with a single string change. ## Prerequisites * Python 3.11+ * An API key for whichever upstream provider you want LiteLLM to route to. The example below uses OpenAI (`OPENAI_API_KEY`). ## Install ```bash theme={null} pip install arize-phoenix-evals litellm pandas ``` ## Configure credentials Set the env var for whichever upstream provider you're targeting. LiteLLM reads the matching env var based on the `/` prefix on the model id: ```bash theme={null} export OPENAI_API_KEY="" # Or, for other providers: # export ANTHROPIC_API_KEY="..." # export MISTRAL_API_KEY="..." # export AWS_ACCESS_KEY_ID="..."; export AWS_SECRET_ACCESS_KEY="..." # bedrock/... ``` See [LiteLLM's provider list](https://docs.litellm.ai/docs/providers) for the full env var map. ## Setup the eval LLM ```python theme={null} # eval_setup.py from phoenix.evals import LLM # The `/` prefix tells LiteLLM which backend to dispatch to # and which env var to read. llm = LLM(provider="litellm", model="openai/gpt-5") ``` Swap `openai/gpt-5` for `anthropic/claude-sonnet-4-6-20250929`, `mistral/mistral-large-latest`, `bedrock/us.anthropic.claude-sonnet-4-6`, `ollama/llama3`, etc. — same evaluator code, different backend. ## Run an evaluation This example builds a hallucination classifier and grades two sample question/answer pairs against a reference. The pattern generalizes: replace the prompt template, choices, and DataFrame columns with whatever metric you want to evaluate. ```python theme={null} # example.py import pandas as pd from phoenix.evals import LLM, create_classifier, evaluate_dataframe llm = LLM(provider="litellm", model="openai/gpt-5") HALLUCINATION_PROMPT = """\ Determine whether the answer below is factually supported by the reference. Reply with exactly one of: factual, hallucinated. Question: {input} Answer: {output} Reference: {reference} """ evaluator = create_classifier( name="hallucination", prompt_template=HALLUCINATION_PROMPT, llm=llm, # `choices` maps each label the LLM may emit to a numeric score. # `direction="maximize"` (the default) means higher score is better. choices={"factual": 1.0, "hallucinated": 0.0}, ) df = pd.DataFrame([ { "input": "What is the capital of France?", "output": "Paris is the capital of France.", "reference": "Paris is the capital and most populous city of France.", }, { "input": "What is the capital of France?", "output": "Berlin is the capital of France.", "reference": "Paris is the capital and most populous city of France.", }, ]) results = evaluate_dataframe(dataframe=df, evaluators=[evaluator]) # `hallucination_score` is a Score row (a dict-like with `score`, `label`, # `explanation`, …) — pull the numeric out for a flat display column. results["score"] = results["hallucination_score"].apply(lambda r: r["score"]) print(results[["input", "output", "score"]].to_string()) ``` ### Expected output ```text wrap theme={null} input output score 0 What is the capital of France? Paris is the capital of France. 1.0 1 What is the capital of France? Berlin is the capital of France. 0.0 ``` The full returned DataFrame also includes `hallucination_execution_details` (status + exceptions + timing) and the original `hallucination_score` column with each evaluator result's full dict (`name`, `score`, `label`, `explanation`, `metadata`, `kind`, `direction`) — useful for surfacing the LLM's reasoning, persisting eval rows back to Arize AX, or filtering retries. ## Troubleshooting * **`401` / `403` from the upstream provider.** Verify the relevant env var is set (`OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, etc.) and matches the `/` prefix on your model id. * **`BadRequestError: LLM Provider NOT provided`.** The model id is missing its provider prefix — LiteLLM needs `openai/gpt-5`, not `gpt-5`. Check the [LiteLLM provider docs](https://docs.litellm.ai/docs/providers) for the exact prefix for your backend. * **All rows return the same label.** Your prompt template isn't differentiating cases. Make sure each row's `{input}`/`{output}`/`{reference}` columns expose enough context for the judge to discriminate, and that `choices` lists every label your prompt asks the LLM to emit. * **Some rows fail with timeout / rate-limit.** Pass `max_retries=` to `evaluate_dataframe(...)` (defaults to 3). For large batches, also pass `initial_per_second_request_rate=...` to `LLM(...)` to throttle. * **Logging results back to Arize AX.** This guide stops at producing the eval DataFrame. To attach those evals to existing spans in an Arize AX project, use [`log_evaluations_sync`](/ax/cookbooks/evaluation/evaluations-quickstart#log-evaluations-back-to-arize) on `arize.Client`. * **Routing through a self-hosted LiteLLM Proxy.** Pass `sync_client_kwargs={"api_base": "https://your-proxy.example.com", "api_key": ""}` to `LLM(...)` to point at a hosted LiteLLM gateway instead of letting the SDK call providers directly. ## Resources