Spaces:
Sleeping
Sleeping
code update for streamlit
Browse files- app.py +49 -107
- requirements.txt +3 -0
app.py
CHANGED
@@ -1,10 +1,8 @@
|
|
1 |
import os
|
2 |
import io
|
3 |
import requests
|
|
|
4 |
from dotenv import load_dotenv
|
5 |
-
from fastapi import FastAPI, HTTPException, UploadFile, File
|
6 |
-
from fastapi.middleware.cors import CORSMiddleware
|
7 |
-
from pydantic import BaseModel
|
8 |
from PyPDF2 import PdfReader
|
9 |
from langchain.text_splitter import CharacterTextSplitter
|
10 |
from langchain.embeddings.huggingface import HuggingFaceEmbeddings
|
@@ -12,27 +10,14 @@ from langchain.vectorstores import FAISS
|
|
12 |
from langchain.chains.question_answering import load_qa_chain
|
13 |
from langchain.llms import HuggingFacePipeline
|
14 |
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
|
15 |
-
import streamlit as st
|
16 |
|
17 |
# Disable WANDB
|
18 |
os.environ['WANDB_DISABLED'] = "true"
|
19 |
|
20 |
# Constants
|
21 |
MODEL_PATH = "/home/lab/halyn/gemma/halyn/paper/models/gemma-2-9b-it"
|
22 |
-
FASTAPI_URL = "http://203.249.64.50:8080" # 서버 주소
|
23 |
-
|
24 |
-
app = FastAPI()
|
25 |
|
26 |
-
#
|
27 |
-
app.add_middleware(
|
28 |
-
CORSMiddleware,
|
29 |
-
allow_origins=["*"], # 모든 출처 허용
|
30 |
-
allow_credentials=True,
|
31 |
-
allow_methods=["*"],
|
32 |
-
allow_headers=["*"],
|
33 |
-
)
|
34 |
-
|
35 |
-
# Global variables to store the knowledge base and QA chain
|
36 |
knowledge_base = None
|
37 |
qa_chain = None
|
38 |
|
@@ -40,7 +25,7 @@ def load_pdf(pdf_file):
|
|
40 |
"""
|
41 |
Load and extract text from a PDF.
|
42 |
Args:
|
43 |
-
pdf_file (str)
|
44 |
Returns:
|
45 |
str: Extracted text from the PDF.
|
46 |
"""
|
@@ -52,9 +37,9 @@ def split_text(text):
|
|
52 |
"""
|
53 |
Split the extracted text into chunks.
|
54 |
Args:
|
55 |
-
text (str)
|
56 |
Returns:
|
57 |
-
list
|
58 |
"""
|
59 |
text_splitter = CharacterTextSplitter(
|
60 |
separator="\n", chunk_size=1000, chunk_overlap=200, length_function=len
|
@@ -65,9 +50,9 @@ def create_knowledge_base(chunks):
|
|
65 |
"""
|
66 |
Create a FAISS knowledge base from text chunks.
|
67 |
Args:
|
68 |
-
chunks (list)
|
69 |
Returns:
|
70 |
-
FAISS: A FAISS knowledge base object
|
71 |
"""
|
72 |
embeddings = HuggingFaceEmbeddings()
|
73 |
return FAISS.from_texts(chunks, embeddings)
|
@@ -76,7 +61,7 @@ def load_model(model_path):
|
|
76 |
"""
|
77 |
Load the HuggingFace model and tokenizer, and create a text-generation pipeline.
|
78 |
Args:
|
79 |
-
model_path (str)
|
80 |
Returns:
|
81 |
pipeline: A HuggingFace pipeline for text generation.
|
82 |
"""
|
@@ -84,56 +69,14 @@ def load_model(model_path):
|
|
84 |
model = AutoModelForCausalLM.from_pretrained(model_path)
|
85 |
return pipeline("text-generation", model=model, tokenizer=tokenizer, max_new_tokens=150, temperature=0.1)
|
86 |
|
87 |
-
|
88 |
-
|
89 |
-
|
|
|
90 |
global qa_chain
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
try:
|
95 |
-
pipe = load_model(MODEL_PATH)
|
96 |
-
llm = HuggingFacePipeline(pipeline=pipe)
|
97 |
-
qa_chain = load_qa_chain(llm, chain_type="stuff")
|
98 |
-
except Exception as e:
|
99 |
-
print(f"Error loading model: {e}")
|
100 |
-
raise HTTPException(status_code=500, detail="Failed to load the language model")
|
101 |
-
|
102 |
-
@app.post("/upload_pdf")
|
103 |
-
async def upload_pdf(file: UploadFile = File(...)):
|
104 |
-
global knowledge_base
|
105 |
-
try:
|
106 |
-
contents = await file.read()
|
107 |
-
pdf_file = io.BytesIO(contents)
|
108 |
-
text = load_pdf(pdf_file)
|
109 |
-
chunks = split_text(text)
|
110 |
-
knowledge_base = create_knowledge_base(chunks)
|
111 |
-
return {"message": "PDF uploaded and processed successfully"}
|
112 |
-
except Exception as e:
|
113 |
-
raise HTTPException(status_code=400, detail=f"Failed to process PDF: {str(e)}")
|
114 |
-
|
115 |
-
class Question(BaseModel):
|
116 |
-
text: str
|
117 |
-
|
118 |
-
@app.post("/ask")
|
119 |
-
async def ask_question(question: Question):
|
120 |
-
global knowledge_base, qa_chain
|
121 |
-
if not knowledge_base:
|
122 |
-
raise HTTPException(status_code=400, detail="No PDF has been uploaded yet")
|
123 |
-
if not qa_chain:
|
124 |
-
raise HTTPException(status_code=500, detail="QA chain is not initialized")
|
125 |
-
|
126 |
-
try:
|
127 |
-
docs = knowledge_base.similarity_search(question.text)
|
128 |
-
response = qa_chain.run(input_documents=docs, question=question.text)
|
129 |
-
|
130 |
-
if "Helpful Answer:" in response:
|
131 |
-
response = response.split("Helpful Answer:")[1].strip()
|
132 |
-
|
133 |
-
return {"response": response}
|
134 |
-
except Exception as e:
|
135 |
-
raise HTTPException(status_code=500, detail=f"Error processing question: {str(e)}")
|
136 |
-
|
137 |
|
138 |
# Streamlit UI
|
139 |
def main_page():
|
@@ -146,23 +89,24 @@ def main_page():
|
|
146 |
st.write("Please click the button below.")
|
147 |
|
148 |
if st.button("Click Here :)"):
|
149 |
-
# FastAPI 서버에 PDF 파일 전송
|
150 |
try:
|
151 |
-
|
152 |
-
|
153 |
-
|
154 |
-
|
155 |
-
|
156 |
-
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
|
161 |
-
st.
|
|
|
|
|
|
|
162 |
|
163 |
def chat_page():
|
164 |
-
st.title(f"
|
165 |
-
st.subheader(f"Ask anything about {st.session_state.paper_name}")
|
166 |
|
167 |
if "messages" not in st.session_state:
|
168 |
st.session_state.messages = []
|
@@ -170,37 +114,40 @@ def chat_page():
|
|
170 |
for message in st.session_state.messages:
|
171 |
with st.chat_message(message["role"]):
|
172 |
st.markdown(message["content"])
|
173 |
-
|
174 |
-
if prompt := st.chat_input("Chat here
|
175 |
-
# Add user message to chat history
|
176 |
st.session_state.messages.append({"role": "user", "content": prompt})
|
177 |
|
178 |
-
# Display user message in chat message container
|
179 |
with st.chat_message("user"):
|
180 |
st.markdown(prompt)
|
181 |
|
182 |
-
|
183 |
-
response = get_response_from_fastapi(prompt)
|
184 |
|
185 |
-
# Display assistant response in chat message container
|
186 |
with st.chat_message("assistant"):
|
187 |
st.markdown(response)
|
188 |
|
189 |
-
# Add assistant response to chat history
|
190 |
st.session_state.messages.append({"role": "assistant", "content": response})
|
191 |
|
192 |
if st.button("Go back to main page"):
|
193 |
st.session_state.page = "main"
|
194 |
|
195 |
-
def
|
196 |
try:
|
197 |
-
|
198 |
-
if
|
199 |
-
return
|
200 |
-
|
201 |
-
return
|
202 |
-
|
203 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
204 |
|
205 |
# Streamlit - 초기 페이지 설정
|
206 |
if "page" not in st.session_state:
|
@@ -215,8 +162,3 @@ if st.session_state.page == "main":
|
|
215 |
main_page()
|
216 |
elif st.session_state.page == "chat":
|
217 |
chat_page()
|
218 |
-
|
219 |
-
# FastAPI 앱 실행을 위한 코드
|
220 |
-
if __name__ == "__main__":
|
221 |
-
import uvicorn
|
222 |
-
uvicorn.run(app, host="0.0.0.0", port=8050)
|
|
|
1 |
import os
|
2 |
import io
|
3 |
import requests
|
4 |
+
import streamlit as st
|
5 |
from dotenv import load_dotenv
|
|
|
|
|
|
|
6 |
from PyPDF2 import PdfReader
|
7 |
from langchain.text_splitter import CharacterTextSplitter
|
8 |
from langchain.embeddings.huggingface import HuggingFaceEmbeddings
|
|
|
10 |
from langchain.chains.question_answering import load_qa_chain
|
11 |
from langchain.llms import HuggingFacePipeline
|
12 |
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
|
|
|
13 |
|
14 |
# Disable WANDB
|
15 |
os.environ['WANDB_DISABLED'] = "true"
|
16 |
|
17 |
# Constants
|
18 |
MODEL_PATH = "/home/lab/halyn/gemma/halyn/paper/models/gemma-2-9b-it"
|
|
|
|
|
|
|
19 |
|
20 |
+
# Global variables
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
21 |
knowledge_base = None
|
22 |
qa_chain = None
|
23 |
|
|
|
25 |
"""
|
26 |
Load and extract text from a PDF.
|
27 |
Args:
|
28 |
+
pdf_file (str): The PDF file.
|
29 |
Returns:
|
30 |
str: Extracted text from the PDF.
|
31 |
"""
|
|
|
37 |
"""
|
38 |
Split the extracted text into chunks.
|
39 |
Args:
|
40 |
+
text (str): The full text extracted from the PDF.
|
41 |
Returns:
|
42 |
+
list: A list of text chunks.
|
43 |
"""
|
44 |
text_splitter = CharacterTextSplitter(
|
45 |
separator="\n", chunk_size=1000, chunk_overlap=200, length_function=len
|
|
|
50 |
"""
|
51 |
Create a FAISS knowledge base from text chunks.
|
52 |
Args:
|
53 |
+
chunks (list): A list of text chunks.
|
54 |
Returns:
|
55 |
+
FAISS: A FAISS knowledge base object.
|
56 |
"""
|
57 |
embeddings = HuggingFaceEmbeddings()
|
58 |
return FAISS.from_texts(chunks, embeddings)
|
|
|
61 |
"""
|
62 |
Load the HuggingFace model and tokenizer, and create a text-generation pipeline.
|
63 |
Args:
|
64 |
+
model_path (str): The path to the pre-trained model.
|
65 |
Returns:
|
66 |
pipeline: A HuggingFace pipeline for text generation.
|
67 |
"""
|
|
|
69 |
model = AutoModelForCausalLM.from_pretrained(model_path)
|
70 |
return pipeline("text-generation", model=model, tokenizer=tokenizer, max_new_tokens=150, temperature=0.1)
|
71 |
|
72 |
+
def setup_qa_chain():
|
73 |
+
"""
|
74 |
+
Set up the question-answering chain.
|
75 |
+
"""
|
76 |
global qa_chain
|
77 |
+
pipe = load_model(MODEL_PATH)
|
78 |
+
llm = HuggingFacePipeline(pipeline=pipe)
|
79 |
+
qa_chain = load_qa_chain(llm, chain_type="stuff")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
80 |
|
81 |
# Streamlit UI
|
82 |
def main_page():
|
|
|
89 |
st.write("Please click the button below.")
|
90 |
|
91 |
if st.button("Click Here :)"):
|
|
|
92 |
try:
|
93 |
+
# PDF 파일 처리
|
94 |
+
contents = paper.read()
|
95 |
+
pdf_file = io.BytesIO(contents)
|
96 |
+
text = load_pdf(pdf_file)
|
97 |
+
chunks = split_text(text)
|
98 |
+
global knowledge_base
|
99 |
+
knowledge_base = create_knowledge_base(chunks)
|
100 |
+
|
101 |
+
st.success("PDF successfully processed! You can now ask questions.")
|
102 |
+
st.session_state.paper_name = paper.name[:-4]
|
103 |
+
st.session_state.page = "chat"
|
104 |
+
setup_qa_chain()
|
105 |
+
except Exception as e:
|
106 |
+
st.error(f"Failed to process the PDF: {str(e)}")
|
107 |
|
108 |
def chat_page():
|
109 |
+
st.title(f"Ask anything about {st.session_state.paper_name}")
|
|
|
110 |
|
111 |
if "messages" not in st.session_state:
|
112 |
st.session_state.messages = []
|
|
|
114 |
for message in st.session_state.messages:
|
115 |
with st.chat_message(message["role"]):
|
116 |
st.markdown(message["content"])
|
117 |
+
|
118 |
+
if prompt := st.chat_input("Chat here!"):
|
|
|
119 |
st.session_state.messages.append({"role": "user", "content": prompt})
|
120 |
|
|
|
121 |
with st.chat_message("user"):
|
122 |
st.markdown(prompt)
|
123 |
|
124 |
+
response = get_response_from_model(prompt)
|
|
|
125 |
|
|
|
126 |
with st.chat_message("assistant"):
|
127 |
st.markdown(response)
|
128 |
|
|
|
129 |
st.session_state.messages.append({"role": "assistant", "content": response})
|
130 |
|
131 |
if st.button("Go back to main page"):
|
132 |
st.session_state.page = "main"
|
133 |
|
134 |
+
def get_response_from_model(prompt):
|
135 |
try:
|
136 |
+
global knowledge_base, qa_chain
|
137 |
+
if not knowledge_base:
|
138 |
+
return "No PDF has been uploaded yet."
|
139 |
+
if not qa_chain:
|
140 |
+
return "QA chain is not initialized."
|
141 |
+
|
142 |
+
docs = knowledge_base.similarity_search(prompt)
|
143 |
+
response = qa_chain.run(input_documents=docs, question=prompt)
|
144 |
+
|
145 |
+
if "Helpful Answer:" in response:
|
146 |
+
response = response.split("Helpful Answer:")[1].strip()
|
147 |
+
|
148 |
+
return response
|
149 |
+
except Exception as e:
|
150 |
+
return f"Error: {str(e)}"
|
151 |
|
152 |
# Streamlit - 초기 페이지 설정
|
153 |
if "page" not in st.session_state:
|
|
|
162 |
main_page()
|
163 |
elif st.session_state.page == "chat":
|
164 |
chat_page()
|
|
|
|
|
|
|
|
|
|
requirements.txt
CHANGED
@@ -1,3 +1,6 @@
|
|
1 |
streamlit
|
2 |
requests
|
3 |
PyPDF2
|
|
|
|
|
|
|
|
1 |
streamlit
|
2 |
requests
|
3 |
PyPDF2
|
4 |
+
dotenv
|
5 |
+
langchain
|
6 |
+
transformers
|