File size: 5,623 Bytes
54eae24
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
import os
import chainlit as cl
from dotenv import load_dotenv
from operator import itemgetter
from langchain_huggingface import HuggingFaceEndpoint
from langchain_community.document_loaders import TextLoader
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import FAISS
from langchain_huggingface import HuggingFaceEndpointEmbeddings
from langchain_core.prompts import PromptTemplate
from langchain.schema.output_parser import StrOutputParser
from langchain.schema.runnable import RunnablePassthrough
from langchain.schema.runnable.config import RunnableConfig
from tqdm.asyncio import tqdm_asyncio
import asyncio
from tqdm.asyncio import tqdm

load_dotenv()

HF_LLM_ENDPOINT = os.environ["HF_LLM_ENDPOINT"]
HF_EMBED_ENDPOINT = os.environ["HF_EMBED_ENDPOINT"]
HF_TOKEN = os.environ["HF_TOKEN"]

if not all([HF_LLM_ENDPOINT, HF_EMBED_ENDPOINT, HF_TOKEN]):
    raise ValueError("Missing required Hugging Face API environment variables.")

if not os.path.exists("./data/paul_graham_essays.txt"):
    raise FileNotFoundError("The specified document file is missing.")


document_loader = TextLoader("./data/paul_graham_essays.txt")
documents = document_loader.load()

text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=30)
split_documents = text_splitter.split_documents(documents)

hf_embeddings = HuggingFaceEndpointEmbeddings(
    model=HF_EMBED_ENDPOINT,
    task="feature-extraction",
    huggingfacehub_api_token=HF_TOKEN,
)

async def add_documents_async(vectorstore, documents):
    await vectorstore.aadd_documents(documents)

async def process_batch(vectorstore, batch, is_first_batch, pbar):
    if is_first_batch:
        result = await FAISS.afrom_documents(batch, hf_embeddings)
    else:
        await add_documents_async(vectorstore, batch)
        result = vectorstore
    pbar.update(len(batch))
    return result

async def main():
    print("Indexing Files")
    
    vectorstore = None
    batch_size = 32
    
    batches = [split_documents[i:i+batch_size] for i in range(0, len(split_documents), batch_size)]
    
    async def process_all_batches():
        nonlocal vectorstore
        tasks = []
        pbars = []
        
        for i, batch in enumerate(batches):
            pbar = tqdm(total=len(batch), desc=f"Batch {i+1}/{len(batches)}", position=i)
            pbars.append(pbar)
            
            if i == 0:
                vectorstore = await process_batch(None, batch, True, pbar)
            else:
                tasks.append(process_batch(vectorstore, batch, False, pbar))
        
        if tasks:
            await asyncio.gather(*tasks)
        
        for pbar in pbars:
            pbar.close()
    
    await process_all_batches()
    
    hf_retriever = vectorstore.as_retriever()
    print("\nIndexing complete. Vectorstore is ready for use.")
    return hf_retriever

async def run():
    retriever = await main()
    return retriever

hf_retriever = asyncio.run(run())


RAG_PROMPT_TEMPLATE = """\
<|start_header_id|>system<|end_header_id|>
You are a helpful assistant. You answer user questions based on provided context. If you can't answer the question with the provided context, say you don't know.<|eot_id|>

<|start_header_id|>user<|end_header_id|>
User Query:
{query}

Context:
{context}<|eot_id|>

<|start_header_id|>assistant<|end_header_id|>
"""

rag_prompt = PromptTemplate.from_template(RAG_PROMPT_TEMPLATE)


### 1. CREATE HUGGINGFACE ENDPOINT FOR LLM
hf_llm = HuggingFaceEndpoint(
    endpoint_url=HF_LLM_ENDPOINT,
    max_new_tokens=512,
    top_k=10,
    top_p=0.95,
    temperature=0.3,
    repetition_penalty=1.15,
    huggingfacehub_api_token=HF_TOKEN,
)


@cl.on_chat_start
async def start_chat():
    """
    This function will be called at the start of every user session. 

    We will build our LCEL RAG chain here, and store it in the user session. 

    The user session is a dictionary that is unique to each user session, and is stored in the memory of the server.
    """

    ### BUILD LCEL RAG CHAIN THAT ONLY RETURNS TEXT
    lcel_rag_chain = (
        {"context": itemgetter("query") | hf_retriever, "query": itemgetter("query")}
        | rag_prompt | hf_llm
    )

    cl.user_session.set("lcel_rag_chain", lcel_rag_chain)
    await cl.Message(content="Welcome to the Paul Graham Essay Bot! Ask me anything about Paul Graham's essays.").send()

@cl.on_message  
async def main(message: cl.Message):
    """
    This function will be called every time a message is received from a session.

    We will use the LCEL RAG chain to generate a response to the user query.

    The LCEL RAG chain is stored in the user session, and is unique to each user session - this is why we can access it here.
    """
    lcel_rag_chain = cl.user_session.get("lcel_rag_chain")

    msg = cl.Message(content="")

    # Create a buffer to store the complete response
    complete_response = ""

    async for chunk in lcel_rag_chain.astream(
        {"query": message.content},
        config=RunnableConfig(callbacks=[cl.LangchainCallbackHandler()]),
    ):
        # Clean the chunk of any <|eot_id|> tokens
        clean_chunk = chunk.replace("<|eot_id|>", "")
        complete_response += chunk  # Store original for final cleaning
        await msg.stream_token(clean_chunk)

    # If <|eot_id|> appears at the end of the complete response, remove it
    if complete_response.endswith("<|eot_id|>"):
        # Update the final message content without the token
        await msg.update(content=complete_response.replace("<|eot_id|>", ""))

    await msg.send()