File size: 3,829 Bytes
b3d4017
 
 
 
 
c97999e
91e3a88
c97999e
 
eaf9631
09a402e
2fa94b3
ba66f78
b3d4017
09a402e
 
 
 
 
 
 
 
 
 
 
b3d4017
c97999e
 
 
b3d4017
c97999e
 
 
 
 
 
 
 
 
 
09a402e
 
c97999e
 
 
 
 
 
 
 
2fa94b3
 
91e3a88
82d14b3
 
2fa94b3
 
 
 
 
 
 
 
 
 
82d14b3
c97999e
 
 
b3d4017
b31d773
82d14b3
09a402e
 
 
82d14b3
2fa94b3
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
from llama_index.core.agent.workflow import AgentWorkflow
from llama_index.core.workflow import Context
from llama_index.core.tools import FunctionTool
from llama_index.llms.huggingface_api import HuggingFaceInferenceAPI
from llama_index.tools.duckduckgo import DuckDuckGoSearchToolSpec
from llama_index.tools.wikipedia import WikipediaToolSpec
from llama_index.core.tools.tool_spec.load_and_search import LoadAndSearchToolSpec
from llama_index.readers.web import SimpleWebPageReader
from llama_index.core.tools.ondemand_loader_tool import OnDemandLoaderTool
from langfuse.llama_index import LlamaIndexInstrumentor
from llama_index.llms.ollama import Ollama
from llama_index.core.agent.workflow import ReActAgent, FunctionAgent

class BasicAgent:
    def __init__(self, ollama=False, langfuse=True):
        if not ollama:
            llm = HuggingFaceInferenceAPI(model_name="Qwen/Qwen2.5-Coder-32B-Instruct")
        else:
            llm = Ollama(model="mistral:latest", request_timeout=120.0)

        # Langfuse
        self.langfuse = langfuse
        if self.langfuse:
            self.instrumentor = LlamaIndexInstrumentor()
            self.instrumentor.start()

        # Initialize tools
        tool_spec = DuckDuckGoSearchToolSpec()
        search_tool = FunctionTool.from_defaults(tool_spec.duckduckgo_full_search)

        wiki_spec = WikipediaToolSpec()
        wiki_search_tool = wiki_spec.to_tool_list()[1]

        # Convert into a LoadAndSearchToolSpec because the wikipedia search tool returns
        # entire Wikipedia pages and this can pollute the context window of the LLM
        wiki_spec = WikipediaToolSpec()
        wiki_search_tool = wiki_spec.to_tool_list()[1]

        # Convert into a LoadAndSearchToolSpec because the wikipedia search tool returns
        # entire Wikipedia pages and this can pollute the context window of the LLM

        # TODO this does not work so well. We need to make the retriever return the top 5 chunks or sth.
        wiki_search_tool_las = LoadAndSearchToolSpec.from_defaults(wiki_search_tool).to_tool_list()

        webpage_tool = OnDemandLoaderTool.from_defaults(
            SimpleWebPageReader(html_to_text=True),
            name="Webpage search tool",
            description="A tool for loading the content of a webpage and querying it for information",
        )

        self.agent = AgentWorkflow.from_tools_or_functions( # ReActAgent(
            tools=[search_tool], # webpage_tool does not work properly - cookies etc
            llm=llm,
            verbose=True,
            system_prompt = (
                "You are a general AI assistant. I will ask you a question. "
                "Report your thoughts, and finish your answer with the following template: "
                "FINAL ANSWER: [YOUR FINAL ANSWER]. YOUR FINAL ANSWER should be a number "
                "OR as few words as possible OR a comma separated list of numbers and/or "
                "strings. If you are asked for a number, don't use comma to write your "
                "number neither use units such as $ or percent sign unless specified otherwise. "
                "If you are asked for a string, don't use articles, neither abbreviations (e.g. "
                "for cities), and write the digits in plain text unless specified otherwise. If "
                "you are asked for a comma separated list, apply the above rules depending of "
                "whether the element to be put in the list is a number or a string."
            )
        )

        # self.ctx = Context(self.agent)

    async def __call__(self, question: str) -> str:
        response = await self.agent.run(user_msg=question) # ctx=self.ctx)

        if self.langfuse:
            self.instrumentor.flush()

        return response.response.content.replace("FINAL ANSWER:", "").strip()