Skip to main content

Nerif Agent

Agent Class​

SimpleChatAgent​

A simple agent class for the Nerif project. This class implements a simple chat agent for the Nerif project. It uses OpenAI's GPT models to generate responses to user inputs.

Attributes:

  • proxy_url (str): The URL of the proxy server for API requests.
  • api_key (str): The API key for authentication.
  • model (str): The name of the GPT model to use.
  • default_prompt (str): The default system prompt for the chat.
  • temperature (float): The temperature setting for response generation.
  • counter (NerifTokenCounter): Token counter instance.
  • messages (List[Any]): The conversation history.
  • max_tokens (int): The maximum number of tokens to generate in the response.

Methods:

  • reset(prompt=None): Resets the conversation history.
  • set_max_tokens(max_tokens=None|int): Sets the maximum tokens limit.
  • chat(message, append=False, max_tokens=None|int): Sends a message and gets a response.

Init:

    def __init__(
        self,
        proxy_url: Optional[str] = None,
        api_key: Optional[str] = None,
        model: str = NERIF_DEFAULT_LLM_MODEL,
        default_prompt: str = "You are a helpful assistant. You can help me by answering my questions.",
        temperature: float = 0.0,
        counter: NerifTokenCounter = None,
        max_tokens: None | int = None,
    )

Example:

import nerif

agent = nerif.agent.SimpleChatAgent()

print(agent.chat("What is the capital of the moon?"))
print(agent.chat("What is the capital of the moon?", max_tokens=10))

SimpleEmbeddingAgent​

A simple agent class for embedding text in the Nerif project. This class provides text embedding functionality using OpenAI's embedding models. It converts text strings into numerical vector representations.

Attributes:

  • proxy_url (str): The URL of the proxy server for API requests.
  • api_key (str): The API key for authentication.
  • model (str): The name of the embedding model to use.
  • counter (NerifTokenCounter): Optional counter for tracking token usage.

Methods:

  • encode(string: str) -> List[float]: Encodes a string into an embedding vector.

Init:

def __init__(
    self,
    proxy_url: Optional[str] = None,
    # base_url: Optional[str] = None,
    api_key: Optional[str] = None,
    model: str = "text-embedding-3-small",
    counter: Optional[NerifTokenCounter] = None,
):

Example:

import nerif

embedding_agent = nerif.agent.SimpleEmbeddingAgent()
print(embedding_agent.encode("What is the capital of the moon?"))

LogitsAgent​

A simple agent for fetching logits from a model. This class provides functionality to get logit probabilities along with model responses.

Attributes:

  • proxy_url (str): The URL of the proxy server for API requests.
  • api_key (str): The API key for authentication.
  • model (str): The name of the model to use.
  • default_prompt (str): The default system prompt for the chat.
  • temperature (float): The temperature setting for response generation.
  • counter (NerifTokenCounter): Token counter instance.
  • messages (List[Any]): The conversation history.
  • max_tokens (int): The maximum number of tokens to generate in the response.

Methods:

  • reset(): Resets the conversation history.
  • set_max_tokens(max_tokens: None|int): Sets the maximum tokens limit.
  • chat(message: str, max_tokens: None|int, logprobs: bool = True, top_logprobs: int = 5) -> Any: Sends a message and gets a response with logits. The logprobs parameter enables logit probabilities, and top_logprobs controls how many top probabilities to return.

Init:

def __init__(
    self,
    proxy_url: Optional[str] = None,
    api_key: Optional[str] = None,
    model: str = NERIF_DEFAULT_LLM_MODEL,
    default_prompt: str = "You are a helpful assistant. You can help me by answering my questions.",
    temperature: float = 0.0,
    counter: Optional[NerifTokenCounter] = None,
    max_tokens: int | None = None,
):

Example:

import nerif

logits_agent = nerif.agent.LogitsAgent()
print(logits_agent.chat("What is the capital of the moon?"))

VisionAgent​

A simple agent for vision tasks in the Nerif project. This class implements a vision-capable agent that can process both text and images. It uses OpenAI's GPT-4 Vision models to generate responses to user inputs that include images.

Attributes:

  • proxy_url (str): The URL of the proxy server for API requests.
  • api_key (str): The API key for authentication.
  • model (str): The name of the GPT model to use.
  • default_prompt (str): The default system prompt for the chat.
  • temperature (float): The temperature setting for response generation.
  • counter (NerifTokenCounter): Token counter instance.
  • max_tokens (int): Maximum tokens to generate in responses.

Methods:

  • append_message(message_type, content): Adds an image or text message to the conversation.
  • reset(): Resets the conversation history.
  • set_max_tokens(max_tokens): Sets the maximum response token length.
  • chat(input: List[Any], append: bool, max_tokens: int|None) -> str: Processes the input and returns a response.

Init:

def __init__(
    self,
    proxy_url: Optional[str] = None,
    api_key: Optional[str] = None,
    model: str = NERIF_DEFAULT_LLM_MODEL,
    default_prompt: str = "You are a helpful assistant. You can help me by answering my questions.",
    temperature: float = 0.0,
    counter: Optional[NerifTokenCounter] = None,
    max_tokens: int | None = None,
):

Example:

from nerif.agent import MessageType, VisionAgent

if __name__ == "__main__":
    vision_model = VisionAgent(model="openrouter/openai/gpt-4o-2024-08-06")
    vision_model.append_message(
        MessageType.IMAGE_URL,
        "https://www.google.com/images/branding/googlelogo/1x/googlelogo_color_272x92dp.png",
    )
    vision_model.append_message(MessageType.TEXT, "what is in this image?")
    response = vision_model.chat()
    print(response)