deep-searcher/deepsearcher/llm/base.py

import ast
import re
from abc import ABC


class BaseLLM(ABC):
    """
    Abstract base class for language model implementations.

    This class defines the interface for language model implementations,
    including methods for chat-based interactions and parsing responses.
    """

    def __init__(self):
        """
        Initialize a BaseLLM object.
        """
        pass

    def chat(self, messages: list[dict]) -> str:
        """
        Send a chat message to the language model and get a response.

        Args:
            messages:
                A list of message dictionaries, typically in the format
                [{"role": "system", "content": "..."}, {"role": "user", "content": "..."}]

        Returns:
            response (str)
                The content of the llm response.
        """
        pass

    @staticmethod
    def literal_eval(response: str) -> str:
        """
        Parse a string response into a Python object using ast.literal_eval.

        This method attempts to extract and parse JSON or Python literals from the response content,
        handling various formats like code blocks and special tags.

        Args:
            response: The string content to parse.

        Returns:
            The processed and parsed Python object.

        Raises:
            ValueError: If the response content cannot be parsed.
        """
        response = response.strip()

        response = BaseLLM.remove_think(response)

        try:
            if response.startswith("```") and response.endswith("```"):
                if response.startswith("```python"):
                    response = response[9:-3]
                elif response.startswith("```json"):
                    response = response[7:-3]
                elif response.startswith("```str"):
                    response = response[6:-3]
                elif response.startswith("```\n"):
                    response = response[4:-3]
                else:
                    raise ValueError("Invalid code block format")
            result = ast.literal_eval(response.strip())
        except Exception:
            matches = re.findall(r"(\[.*?\]|\{.*?\})", response, re.DOTALL)

            if len(matches) != 1:
                raise ValueError(
                    f"Invalid JSON/List format for response content:\n{response}"
                )

            json_part = matches[0]
            return ast.literal_eval(json_part)

        return result

    @staticmethod
    def remove_think(response: str) -> str:
        # remove content between <think> and </think>, especial for reasoning model
        if "<think>" in response and "</think>" in response:
            end_of_think = response.find("</think>") + len("</think>")
            response = response[end_of_think:]
        return response.strip()