File size: 4,944 Bytes
1ab26fb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
import gradio as gr
import llama_index
import openai
import phoenix as px
from llama_index import ServiceContext, VectorStoreIndex
from llama_index import set_global_service_context
from llama_index.agent import OpenAIAgent
from llama_index.chat_engine.types import ChatMode
from llama_index.ingestion import IngestionPipeline
from llama_index.tools import QueryEngineTool
from llama_index.vector_stores.qdrant import QdrantVectorStore

from chatbot import Chatbot, IndexBuilder
from constants import CHAT_TEXT_QA_PROMPT, TEXT_QA_SYSTEM_PROMPT, CHUNK_SIZE, DEFAULT_MODEL, \
    IS_LOAD_FROM_VECTOR_STORE
from environments import OPENAI_API_KEY, QDRANT_COLLECTION_NAME
from qdrant import client as qdrant_client
from service_provider_config import get_service_provider_config

px.launch_app()
llama_index.set_global_handler("arize_phoenix")
openai.api_key = OPENAI_API_KEY

llm, embedding_model = get_service_provider_config(model_name=DEFAULT_MODEL)
service_context = ServiceContext.from_defaults(
    chunk_size=CHUNK_SIZE,
    llm=llm,
    embed_model=embedding_model,
)
set_global_service_context(service_context)


class KipIndexBuilder(IndexBuilder):
    def _load_documents(self):
        # TODO: implement logic to import documents into qdrant - API feeding logic to consider
        pass

    def _setup_service_context(self):
        super()._setup_service_context()

    def _setup_vector_store(self):
        self.vector_store = QdrantVectorStore(
            client=qdrant_client, collection_name=self.vdb_collection_name)
        super()._setup_vector_store()

    def _setup_index(self):
        super()._setup_index()
        if self.is_load_from_vector_store:
            self.index = VectorStoreIndex.from_vector_store(self.vector_store)
            print("set up index from vector store")
            return
        pipeline = IngestionPipeline(
            transformations=[
                self.embed_model,
            ],
            vector_store=self.vector_store,
        )
        pipeline.run(documents=self.documents, show_progress=True)
        self.index = VectorStoreIndex.from_vector_store(self.vector_store)


class KipToolChatbot(Chatbot):
    DENIED_ANSWER_PROMPT = ""
    SYSTEM_PROMPT = ""
    CHAT_EXAMPLES = []

    def _setup_observer(self):
        pass

    def _setup_index(self):
        super()._setup_index()

    def _setup_query_engine(self):
        super()._setup_query_engine()
        self.query_engine = self.index.as_query_engine(
            text_qa_template=CHAT_TEXT_QA_PROMPT)

    def _setup_tools(self):
        super()._setup_tools()
        self.tools = QueryEngineTool.from_defaults(
            query_engine=self.query_engine)

    def _setup_chat_engine(self):
        super()._setup_chat_engine()
        self.chat_engine = OpenAIAgent.from_tools(
            tools=[self.tools],
            llm=llm,
            similarity_top_k=1,
            verbose=True
        )


class KipContextChatbot(KipToolChatbot):
    def _setup_chat_engine(self):
        self.chat_engine = self.index.as_chat_engine(
            chat_mode=ChatMode.CONTEXT,
            similarity_top_k=5,
            system_prompt=TEXT_QA_SYSTEM_PROMPT.content,
            text_qa_template=CHAT_TEXT_QA_PROMPT)


class KipSimpleChatbot(KipToolChatbot):
    def _setup_chat_engine(self):
        self.chat_engine = self.index.as_chat_engine(
            chat_mode=ChatMode.SIMPLE)


index_builder = KipIndexBuilder(vdb_collection_name=QDRANT_COLLECTION_NAME,
                                embed_model=embedding_model,
                                is_load_from_vector_store=IS_LOAD_FROM_VECTOR_STORE)

kip_chatbot = KipToolChatbot(model_name=DEFAULT_MODEL, index_builder=index_builder)
kip_chatbot_context = KipContextChatbot(model_name=DEFAULT_MODEL, index_builder=index_builder)
kip_chatbot_simple = KipSimpleChatbot(model_name=DEFAULT_MODEL, index_builder=index_builder)


def vote(data: gr.LikeData):
    if data.liked:
        gr.Info("You up-voted this response: " + data.value)
    else:
        gr.Info("You down-voted this response: " + data.value)


chatbot = gr.Chatbot()

with gr.Blocks() as demo:
    gr.Markdown("# Kipwise LLM demo")

    with gr.Tab("Using relevant context sent to system prompt"):
        context_interface = gr.ChatInterface(
            kip_chatbot_context.stream_chat,
            examples=kip_chatbot.CHAT_EXAMPLES,
        )
        chatbot.like(vote, None, None)

    with gr.Tab("Using function calling as tool to retrieve"):
        function_call_interface = gr.ChatInterface(
            kip_chatbot.stream_chat,
            examples=kip_chatbot.CHAT_EXAMPLES,
        )
        chatbot.like(vote, None, None)

    with gr.Tab("Vanilla ChatGPT without modification"):
        vanilla_interface = gr.ChatInterface(
            kip_chatbot_simple.stream_chat,
            examples=kip_chatbot.CHAT_EXAMPLES)
demo.queue(False).launch(server_name='0.0.0.0', share=False)