Model Configuration

ROSA supports both the OpenAI API and Azure OpenAI for its language model. Users can configure and pass either a ChatOpenAI or AzureChatOpenAI instance to the ROSA class.

Note

We're also experimenting with the use of local models like ChatOllama (see below)

Here's an overview of how to set up and use these LLMs:

Using ChatOpenAI

To use the standard OpenAI API with the ChatOpenAI model:

1. Ensure you have your OpenAI API key.

2. Set up your environment variable:

   Add the following to your .env file or set it in your system environment:

   OPENAI_API_KEY=your_openai_api_key
   

3. Create a ChatOpenAI instance:

   import os
   from dotenv import load_dotenv
   from langchain_openai import ChatOpenAI
   
   load_dotenv()  # This loads the variables from .env file
   
   openai_llm = ChatOpenAI(
       model_name="gpt-4",  # or your preferred model
       temperature=0,
       max_tokens=None,
       timeout=None,
       max_retries=2,
       openai_api_key=os.getenv("OPENAI_API_KEY"),  # Using environment variable
   )
   
   # Pass the LLM to ROSA
   rosa_instance = ROSA(ros_version=2, llm=openai_llm, ...)

Important

Ensure that you have the necessary environment variables set in your .env file or system environment. Always handle your API keys and secrets securely.

Using AzureChatOpenAI

To use Azure OpenAI, you'll need to create an AzureChatOpenAI instance with the appropriate configuration. There are two ways to set this up:

1. Using Azure API Management (APIM) with Tenant ID, Client ID, and Client Secret:

   Required Environment Variables:

   • APIM_SUBSCRIPTION_KEY (if required by your APIM setup)
   • AZURE_TENANT_ID
   • AZURE_CLIENT_ID
   • AZURE_CLIENT_SECRET
   • DEPLOYMENT_ID
   • API_VERSION
   • API_ENDPOINT

   Add these to your .env file or set them in your system environment.

   import os
   from dotenv import load_dotenv
   from langchain_openai import AzureChatOpenAI
   from azure.identity import ClientSecretCredential, get_bearer_token_provider
   
   load_dotenv()
   
   # Set up Azure authentication
   credential = ClientSecretCredential(
       tenant_id=os.getenv("AZURE_TENANT_ID"),
       client_id=os.getenv("AZURE_CLIENT_ID"),
       client_secret=os.getenv("AZURE_CLIENT_SECRET"),
       authority="https://login.microsoftonline.com",
   )
   
   token_provider = get_bearer_token_provider(
       credential, "https://cognitiveservices.azure.com/.default"
   )
   
   # Create AzureChatOpenAI instance
   azure_llm = AzureChatOpenAI(
       azure_deployment=os.getenv("DEPLOYMENT_ID"),
       azure_ad_token_provider=token_provider,
       openai_api_type="azure_ad",
       api_version=os.getenv("API_VERSION"),
       azure_endpoint=os.getenv("API_ENDPOINT"),
       default_headers={"Ocp-Apim-Subscription-Key": os.getenv("APIM_SUBSCRIPTION_KEY")} if os.getenv("APIM_SUBSCRIPTION_KEY") else {},
   )
   
   # Pass the LLM to ROSA
   rosa = ROSA(ros_version=2, llm=azure_llm, ...)

2. Using API Key:

   Required Environment Variables:

   • AZURE_OPENAI_API_KEY
   • DEPLOYMENT_ID
   • API_ENDPOINT

   Add these to your .env file or set them in your system environment.

   import os
   from dotenv import load_dotenv
   from langchain_openai import AzureChatOpenAI
   
   load_dotenv()
   
   # Create AzureChatOpenAI instance
   azure_llm = AzureChatOpenAI(
       azure_deployment=os.getenv("DEPLOYMENT_ID"),
       openai_api_key=os.getenv("AZURE_OPENAI_API_KEY"),
       azure_endpoint=os.getenv("API_ENDPOINT"),
   )
   
   # Pass the LLM to ROSA
   rosa = ROSA(ros_version=2, llm=azure_llm, ...)

Using ChatOllama

Warning

ChatOllama usage is experimental.

The latest Llama models support tool calling, which means they can theoretically be used for building agents. That said, ROSA was built to be used with state-of-the-art models like gpt-4o, with large context windows, high throughput, and relatively low latency. Achieving similar performance with local models would require a lot of expensive hardware and additional setup and configuration.

We have only experimented with Llama3.1 8b, and it is severely limited. We do not recommend using Llama3.1 8b with ROSA. Additionally, the core ROSA agent requires a context window of at least 4k tokens, so we recommend a minimum context length >= 8192 tokens, regardless of which model you use.

To use Ollama with the ChatOllama model:

1. Ensure you have Ollama installed and running on your system.

2. Create a ChatOllama instance:

   from langchain_ollama import ChatOllama
   
   ollama_llm = ChatOllama(
       model="llama3.1:70b",  # or your preferred model
       temperature=0,
       num_ctx=8192,  # adjust based on your model's context window
   )
   
   # Pass the LLM to ROSA
   rosa_instance = ROSA(ros_version=2, llm=ollama_llm, ...)

   You can customize the ChatOllama instance with additional parameters:

   • base_url: The base URL of your Ollama instance (default is "http://localhost:11434")
   • callback_manager: A callback manager for handling events
   • verbose: Whether to print verbose output

3. If you're running Ollama on a different machine or port, specify the base_url:

   ollama_llm = ChatOllama(
       model="llama3.1",
       base_url="http://your-ollama-server:11434",
       # ... other parameters ...
   )

Note

Ollama runs locally on your machine, so you don't need to set up API keys or environment variables. However, ensure that you have sufficient system resources to run your chosen model.

For more detailed information on using these models, including advanced features like tool calling, streaming, and fine-tuning, refer to the official documentation:

• Azure OpenAI Documentation
• OpenAI Documentation
• Ollama Documentation

Important

Remember to handle your API keys and secrets securely, preferably using environment variables or a secure secret management system.