langchain/templates/retrieval-agent-fireworks/retrieval_agent_fireworks/chain.py

from typing import List

from langchain import hub
from langchain.agents import AgentExecutor
from langchain.agents.format_scratchpad import format_log_to_str
from langchain.agents.output_parsers import ReActJsonSingleInputOutputParser
from langchain.callbacks.manager import CallbackManagerForRetrieverRun
from langchain.schema import BaseRetriever, Document
from langchain.tools.render import render_text_description
from langchain.tools.retriever import create_retriever_tool
from langchain_community.chat_models.fireworks import ChatFireworks
from langchain_community.utilities.arxiv import ArxivAPIWrapper
from langchain_core.pydantic_v1 import BaseModel

MODEL_ID = "accounts/fireworks/models/mixtral-8x7b-instruct"


class ArxivRetriever(BaseRetriever, ArxivAPIWrapper):
    """`Arxiv` retriever.

    It wraps load() to get_relevant_documents().
    It uses all ArxivAPIWrapper arguments without any change.
    """

    get_full_documents: bool = False

    def _get_relevant_documents(
        self, query: str, *, run_manager: CallbackManagerForRetrieverRun
    ) -> List[Document]:
        try:
            if self.is_arxiv_identifier(query):
                results = self.arxiv_search(
                    id_list=query.split(),
                    max_results=self.top_k_results,
                ).results()
            else:
                results = self.arxiv_search(  # type: ignore
                    query[: self.ARXIV_MAX_QUERY_LENGTH], max_results=self.top_k_results
                ).results()
        except self.arxiv_exceptions as ex:
            return [Document(page_content=f"Arxiv exception: {ex}")]
        docs = [
            Document(
                page_content=result.summary,
                metadata={
                    "Published": result.updated.date(),
                    "Title": result.title,
                    "Authors": ", ".join(a.name for a in result.authors),
                },
            )
            for result in results
        ]
        return docs


# Set up tool(s)
description = (
    "A wrapper around Arxiv.org "
    "Useful for when you need to answer questions about Physics, Mathematics, "
    "Computer Science, Quantitative Biology, Quantitative Finance, Statistics, "
    "Electrical Engineering, and Economics "
    "from scientific articles on arxiv.org. "
    "Input should be a search query."
)
arxiv_tool = create_retriever_tool(ArxivRetriever(), "arxiv", description)
tools = [arxiv_tool]

# Set up LLM
llm = ChatFireworks(
    model=MODEL_ID,
    model_kwargs={
        "temperature": 0,
        "max_tokens": 2048,
        "top_p": 1,
    },
    cache=True,
)

# setup ReAct style prompt
prompt = hub.pull("hwchase17/react-json")
prompt = prompt.partial(
    tools=render_text_description(tools),
    tool_names=", ".join([t.name for t in tools]),
)

# define the agent
model_with_stop = llm.bind(stop=["\nObservation"])
agent = (
    {
        "input": lambda x: x["input"],
        "agent_scratchpad": lambda x: format_log_to_str(x["intermediate_steps"]),
    }
    | prompt
    | model_with_stop
    | ReActJsonSingleInputOutputParser()
)


class InputType(BaseModel):
    input: str


# instantiate AgentExecutor
agent_executor = AgentExecutor(
    agent=agent,
    tools=tools,
    verbose=True,
    handle_parsing_errors=True,
).with_types(input_type=InputType)