ThisIs-Developer
commited on
Commit
•
da12358
1
Parent(s):
0dcdc5b
Upload model.py
Browse files- Chainlit/model.py +98 -0
Chainlit/model.py
ADDED
@@ -0,0 +1,98 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import asyncio
|
2 |
+
from langchain.document_loaders import PyPDFLoader, DirectoryLoader
|
3 |
+
from langchain import PromptTemplate
|
4 |
+
from langchain.embeddings import HuggingFaceEmbeddings
|
5 |
+
from langchain.vectorstores import FAISS
|
6 |
+
from langchain.llms import CTransformers
|
7 |
+
from langchain.chains import RetrievalQA
|
8 |
+
import chainlit as cl
|
9 |
+
|
10 |
+
DB_FAISS_PATH = 'vectorstores/db_faiss'
|
11 |
+
|
12 |
+
custom_prompt_template = """Use the following pieces of information to answer the user's question.
|
13 |
+
If you don't know the answer, just say that you don't know, don't try to make up an answer.
|
14 |
+
|
15 |
+
Context: {context}
|
16 |
+
Question: {question}
|
17 |
+
|
18 |
+
Only return the helpful answer below and nothing else.
|
19 |
+
Helpful answer:
|
20 |
+
"""
|
21 |
+
|
22 |
+
def set_custom_prompt():
|
23 |
+
"""
|
24 |
+
Prompt template for QA retrieval for each vectorstore
|
25 |
+
"""
|
26 |
+
prompt = PromptTemplate(template=custom_prompt_template,
|
27 |
+
input_variables=['context', 'question'])
|
28 |
+
return prompt
|
29 |
+
|
30 |
+
# Retrieval QA Chain
|
31 |
+
def retrieval_qa_chain(llm, prompt, db):
|
32 |
+
qa_chain = RetrievalQA.from_chain_type(llm=llm,
|
33 |
+
chain_type='stuff',
|
34 |
+
retriever=db.as_retriever(search_kwargs={'k': 2}),
|
35 |
+
return_source_documents=True,
|
36 |
+
chain_type_kwargs={'prompt': prompt}
|
37 |
+
)
|
38 |
+
return qa_chain
|
39 |
+
|
40 |
+
# Loading the model
|
41 |
+
def load_llm():
|
42 |
+
# Load the locally downloaded model here
|
43 |
+
llm = CTransformers(
|
44 |
+
model="TheBloke/Llama-2-7B-Chat-GGML",
|
45 |
+
model_type="llama",
|
46 |
+
max_new_tokens=512,
|
47 |
+
temperature=0.5
|
48 |
+
)
|
49 |
+
return llm
|
50 |
+
|
51 |
+
# QA Model Function
|
52 |
+
async def qa_bot():
|
53 |
+
embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2",
|
54 |
+
model_kwargs={'device': 'cpu'})
|
55 |
+
db = FAISS.load_local(DB_FAISS_PATH, embeddings)
|
56 |
+
llm = load_llm()
|
57 |
+
qa_prompt = set_custom_prompt()
|
58 |
+
qa = retrieval_qa_chain(llm, qa_prompt, db)
|
59 |
+
|
60 |
+
return qa
|
61 |
+
|
62 |
+
# Output function
|
63 |
+
async def final_result(query):
|
64 |
+
qa_result = await qa_bot()
|
65 |
+
response = await qa_result({'query': query})
|
66 |
+
return response
|
67 |
+
|
68 |
+
# chainlit code
|
69 |
+
@cl.on_chat_start
|
70 |
+
async def start():
|
71 |
+
chain = await qa_bot()
|
72 |
+
# msg = cl.Message(content="Starting the bot...")
|
73 |
+
# await msg.send()
|
74 |
+
# msg.content = "Hi, Welcome to Medical Bot. What is your query?"
|
75 |
+
# await msg.update()
|
76 |
+
|
77 |
+
cl.user_session.set("chain", chain)
|
78 |
+
|
79 |
+
@cl.on_message
|
80 |
+
async def main(message):
|
81 |
+
chain = cl.user_session.get("chain")
|
82 |
+
cb = cl.AsyncLangchainCallbackHandler(
|
83 |
+
stream_final_answer=True, answer_prefix_tokens=["FINAL", "ANSWER"]
|
84 |
+
)
|
85 |
+
cb.answer_reached = True
|
86 |
+
res = await chain.acall(message.content, callbacks=[cb])
|
87 |
+
answer = res["result"]
|
88 |
+
sources = res["source_documents"]
|
89 |
+
|
90 |
+
if sources:
|
91 |
+
answer += f"\nSources:" + str(sources)
|
92 |
+
else:
|
93 |
+
answer += "\nNo sources found"
|
94 |
+
|
95 |
+
await cl.Message(content=answer).send()
|
96 |
+
|
97 |
+
if __name__ == "__main__":
|
98 |
+
asyncio.run(cl.main())
|