Phoenix supports various LLMs for evaluation, including Gemini and Claude, through the LiteLLM library. By default, Phoenix uses GPT models, but you can configure it to use other models.
To use Gemini, Claude, and other LLMs, you can follow these steps:
1. Install LiteLLM: Ensure you have the LiteLLM package installed.
2. Set Environment Variables: Configure the necessary API keys and environment variables for the LLMs you want to use.
3. Use the completion Method: Call the completion method from LiteLLM to interact with the desired LLM.
Here is an example code snippet for using different LLMs:
```
from litellm import completion
os.environ["OPENAI_API_KEY"] = "your_openai_key"
os.environ["COHERE_API_KEY"] = "your_cohere_key"
os.environ["ANTHROPIC_API_KEY"] = "your_anthropic_key"
messages = [{"content": "Hello, how are you?", "role": "user"}]
response_openai = completion(model="gpt-3.5-turbo", messages=messages)
response_cohere = completion(model="command-nightly", messages=messages)
response_anthropic = completion(model="claude-instant-1", messages=messages)
print(response_openai, response_cohere, response_anthropic)
LBKSFor more detailed instructions, you can refer to the [LiteLLM documentation](https://docs.litellm.ai/docs/providers/huggingface) and the [Phoenix API documentation](https://docs.arize.com/phoenix/api/evaluation-models#litellmmodel).
Would you like more detailed steps or a specific example for a particular LLM?
Sources:
- [https://github.com/Arize-ai/phoenix/issues/2140](https://github.com/Arize-ai/phoenix/issues/2140)
- [https://github.com/Arize-ai/phoenix/issues/2117](https://github.com/Arize-ai/phoenix/issues/2117)
