Carlos Salgado commited on
Commit
9813b6b
1 Parent(s): 24dc52a

rename scripts.py, remove unused files

Browse files
Files changed (4) hide show
  1. backend/draft_app.py +219 -0
  2. html_templates.py +0 -44
  3. requirements.txt +1 -4
  4. scripts.py +102 -0
backend/draft_app.py ADDED
@@ -0,0 +1,219 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import time
2
+ import streamlit as st
3
+ from PyPDF2 import PdfReader
4
+ from langchain.text_splitter import CharacterTextSplitter
5
+ from langchain.embeddings import OpenAIEmbeddings, HuggingFaceInstructEmbeddings
6
+ from langchain.vectorstores import FAISS
7
+ from langchain.chat_models import ChatOpenAI
8
+ from langchain.memory import ConversationBufferMemory
9
+ from langchain.chains import ConversationalRetrievalChain
10
+ import os
11
+ import pickle
12
+ from datetime import datetime
13
+ from backend.generate_metadata import generate_metadata, ingest
14
+
15
+
16
+ css = '''
17
+ <style>
18
+ .chat-message {
19
+ padding: 1.5rem; border-radius: 0.5rem; margin-bottom: 1rem; display: flex
20
+ }
21
+ .chat-message.user {
22
+ background-color: #2b313e
23
+ }
24
+ .chat-message.bot {
25
+ background-color: #475063
26
+ }
27
+ .chat-message .avatar {
28
+ width: 20%;
29
+ }
30
+ .chat-message .avatar img {
31
+ max-width: 78px;
32
+ max-height: 78px;
33
+ border-radius: 50%;
34
+ object-fit: cover;
35
+ }
36
+ .chat-message .message {
37
+ width: 80%;
38
+ padding: 0 1.5rem;
39
+ color: #fff;
40
+ }
41
+ '''
42
+ bot_template = '''
43
+ <div class="chat-message bot">
44
+ <div class="avatar">
45
+ <img src="https://i.ibb.co/cN0nmSj/Screenshot-2023-05-28-at-02-37-21.png"
46
+ style="max-height: 78px; max-width: 78px; border-radius: 50%; object-fit: cover;">
47
+ </div>
48
+ <div class="message">{{MSG}}</div>
49
+ </div>
50
+ '''
51
+ user_template = '''
52
+ <div class="chat-message user">
53
+ <div class="avatar">
54
+ <img src="https://i.ibb.co/rdZC7LZ/Photo-logo-1.png">
55
+ </div>
56
+ <div class="message">{{MSG}}</div>
57
+ </div>
58
+ '''
59
+
60
+
61
+ def get_pdf_text(pdf_docs):
62
+ text = ""
63
+ for pdf in pdf_docs:
64
+ pdf_reader = PdfReader(pdf)
65
+ for page in pdf_reader.pages:
66
+ text += page.extract_text()
67
+ return text
68
+
69
+
70
+ def get_text_chunks(text):
71
+ text_splitter = CharacterTextSplitter(
72
+ separator="\n",
73
+ chunk_size=1000,
74
+ chunk_overlap=200,
75
+ length_function=len
76
+ )
77
+ chunks = text_splitter.split_text(text)
78
+ return chunks
79
+
80
+
81
+ def get_vectorstore(text_chunks):
82
+ embeddings = OpenAIEmbeddings()
83
+ # embeddings = HuggingFaceInstructEmbeddings(model_name="hkunlp/instructor-xl")
84
+ vectorstore = FAISS.from_texts(texts=text_chunks, embedding=embeddings)
85
+ return vectorstore
86
+
87
+
88
+ def get_conversation_chain(vectorstore):
89
+ llm = ChatOpenAI()
90
+ # llm = HuggingFaceHub(repo_id="google/flan-t5-xxl", model_kwargs={"temperature":0.5, "max_length":512})
91
+
92
+ memory = ConversationBufferMemory(
93
+ memory_key='chat_history', return_messages=True)
94
+ conversation_chain = ConversationalRetrievalChain.from_llm(
95
+ llm=llm,
96
+ retriever=vectorstore.as_retriever(),
97
+ memory=memory
98
+ )
99
+ return conversation_chain
100
+
101
+
102
+ def handle_userinput(user_question):
103
+ response = st.session_state.conversation({'question': user_question})
104
+ st.session_state.chat_history = response['chat_history']
105
+
106
+ for i, message in enumerate(st.session_state.chat_history):
107
+ # Display user message
108
+ if i % 2 == 0:
109
+ st.write(user_template.replace("{{MSG}}", message.content), unsafe_allow_html=True)
110
+ else:
111
+ print(message)
112
+ # Display AI response
113
+ st.write(bot_template.replace("{{MSG}}", message.content), unsafe_allow_html=True)
114
+
115
+ # THIS DOESNT WORK, SOMEONE PLS FIX
116
+ # Display source document information if available in the message
117
+ if hasattr(message, 'source') and message.source:
118
+ st.write(f"Source Document: {message.source}", unsafe_allow_html=True)
119
+
120
+
121
+ def safe_vec_store():
122
+ # USE VECTARA INSTEAD
123
+ os.makedirs('vectorstore', exist_ok=True)
124
+ filename = 'vectores' + datetime.now().strftime('%Y%m%d%H%M') + '.pkl'
125
+ file_path = os.path.join('vectorstore', filename)
126
+ vector_store = st.session_state.vectorstore
127
+
128
+ # Serialize and save the entire FAISS object using pickle
129
+ with open(file_path, 'wb') as f:
130
+ pickle.dump(vector_store, f)
131
+
132
+
133
+ def main():
134
+ st.set_page_config(page_title="Doc Verify RAG", page_icon=":hospital:")
135
+ st.write(css, unsafe_allow_html=True)
136
+ if "openai_api_key" not in st.session_state:
137
+ st.session_state.openai_api_key = False
138
+ if "openai_org" not in st.session_state:
139
+ st.session_state.openai_org = False
140
+ if "classify" not in st.session_state:
141
+ st.session_state.classify = False
142
+ def set_pw():
143
+ st.session_state.openai_api_key = True
144
+ st.subheader("Your documents")
145
+ # OPENAI_ORG_ID = st.text_input("OPENAI ORG ID:")
146
+ OPENAI_API_KEY = st.text_input("OPENAI API KEY:", type="password",
147
+ disabled=st.session_state.openai_api_key, on_change=set_pw)
148
+ if st.session_state.classify:
149
+ pdf_doc = st.file_uploader("Upload your PDFs here and click on 'Process'", accept_multiple_files=False)
150
+ else:
151
+ pdf_docs = st.file_uploader("Upload your PDFs here and click on 'Process'", accept_multiple_files=True)
152
+ filenames = [file.name for file in pdf_docs if file is not None]
153
+ if st.button("Process"):
154
+ with st.spinner("Processing"):
155
+ if st.session_state.classify:
156
+ # THE CLASSIFICATION APP
157
+ st.write("Classifying")
158
+ plain_text_doc = ingest(pdf_doc.name)
159
+ classification_result = generate_metadata(plain_text_doc)
160
+ st.write(classification_result)
161
+ else:
162
+ # NORMAL RAG
163
+ loaded_vec_store = None
164
+ for filename in filenames:
165
+ if ".pkl" in filename:
166
+ file_path = os.path.join('vectorstore', filename)
167
+ with open(file_path, 'rb') as f:
168
+ loaded_vec_store = pickle.load(f)
169
+ raw_text = get_pdf_text(pdf_docs)
170
+ text_chunks = get_text_chunks(raw_text)
171
+ vec = get_vectorstore(text_chunks)
172
+ if loaded_vec_store:
173
+ vec.merge_from(loaded_vec_store)
174
+ st.warning("loaded vectorstore")
175
+ if "vectorstore" in st.session_state:
176
+ vec.merge_from(st.session_state.vectorstore)
177
+ st.warning("merged to existing")
178
+ st.session_state.vectorstore = vec
179
+ st.session_state.conversation = get_conversation_chain(vec)
180
+ st.success("data loaded")
181
+
182
+
183
+ if "conversation" not in st.session_state:
184
+ st.session_state.conversation = None
185
+ if "chat_history" not in st.session_state:
186
+ st.session_state.chat_history = None
187
+
188
+ st.header("Doc Verify RAG :hospital:")
189
+ user_question = st.text_input("Ask a question about your documents:")
190
+ if user_question:
191
+ handle_userinput(user_question)
192
+ with st.sidebar:
193
+
194
+ st.subheader("Classification Instrucitons")
195
+ classifier_docs = st.file_uploader("Upload your instructions here and click on 'Process'", accept_multiple_files=True)
196
+ filenames = [file.name for file in classifier_docs if file is not None]
197
+
198
+ if st.button("Process Classification"):
199
+ st.session_state.classify = True
200
+ with st.spinner("Processing"):
201
+ st.warning("set classify")
202
+ time.sleep(3)
203
+
204
+
205
+ # Save and Load Embeddings
206
+ if st.button("Save Embeddings"):
207
+ if "vectorstore" in st.session_state:
208
+ safe_vec_store()
209
+ # st.session_state.vectorstore.save_local("faiss_index")
210
+ st.sidebar.success("saved")
211
+ else:
212
+ st.sidebar.warning("No embeddings to save. Please process documents first.")
213
+
214
+ if st.button("Load Embeddings"):
215
+ st.warning("this function is not in use, just upload the vectorstore")
216
+
217
+
218
+ if __name__ == '__main__':
219
+ main()
html_templates.py DELETED
@@ -1,44 +0,0 @@
1
- css = '''
2
- <style>
3
- .chat-message {
4
- padding: 1.5rem; border-radius: 0.5rem; margin-bottom: 1rem; display: flex
5
- }
6
- .chat-message.user {
7
- background-color: #2b313e
8
- }
9
- .chat-message.bot {
10
- background-color: #475063
11
- }
12
- .chat-message .avatar {
13
- width: 20%;
14
- }
15
- .chat-message .avatar img {
16
- max-width: 78px;
17
- max-height: 78px;
18
- border-radius: 50%;
19
- object-fit: cover;
20
- }
21
- .chat-message .message {
22
- width: 80%;
23
- padding: 0 1.5rem;
24
- color: #fff;
25
- }
26
- '''
27
-
28
- bot_template = '''
29
- <div class="chat-message bot">
30
- <div class="avatar">
31
- <img src="" style="max-height: 78px; max-width: 78px; border-radius: 50%; object-fit: cover;">
32
- </div>
33
- <div class="message">{{MSG}}</div>
34
- </div>
35
- '''
36
-
37
- user_template = '''
38
- <div class="chat-message user">
39
- <div class="avatar">
40
- <img src="">
41
- </div>
42
- <div class="message">{{MSG}}</div>
43
- </div>
44
- '''
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
requirements.txt CHANGED
@@ -5,7 +5,4 @@ unstructured[local-inference]
5
  python-dotenv
6
  streamlit
7
  langchain
8
- openai
9
- chromadb
10
- tiktoken
11
- python-poppler
 
5
  python-dotenv
6
  streamlit
7
  langchain
8
+ openai
 
 
 
scripts.py ADDED
@@ -0,0 +1,102 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import os
2
+ import io
3
+ import argparse
4
+ import json
5
+ import openai
6
+ import sys
7
+ from dotenv import load_dotenv
8
+ from langchain_community.document_loaders import TextLoader
9
+ from langchain_community.document_loaders import UnstructuredPDFLoader
10
+ from langchain_community.embeddings.fake import FakeEmbeddings
11
+ from langchain_text_splitters import RecursiveCharacterTextSplitter
12
+
13
+ load_dotenv()
14
+
15
+
16
+ import io
17
+
18
+ def ingest(file_obj, file_ext='pdf'):
19
+ if file_ext == 'pdf':
20
+ loader = UnstructuredPDFLoader(file_obj)
21
+ elif file_ext == 'txt':
22
+ loader = TextLoader(file_obj)
23
+ else:
24
+ raise NotImplementedError('Only .txt or .pdf files are supported')
25
+
26
+ # transform locally
27
+ documents = loader.load()
28
+ text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=0,
29
+ separators=[
30
+ "\n\n",
31
+ "\n",
32
+ " ",
33
+ ",",
34
+ "\uff0c", # Fullwidth comma
35
+ "\u3001", # Ideographic comma
36
+ "\uff0e", # Fullwidth full stop
37
+ # "\u200B", # Zero-width space (Asian languages)
38
+ # "\u3002", # Ideographic full stop (Asian languages)
39
+ "",
40
+ ])
41
+ docs = text_splitter.split_documents(documents)
42
+
43
+ return docs
44
+
45
+
46
+ def generate_metadata(docs):
47
+ prompt_template = """
48
+ BimDiscipline = ['plumbing', 'network', 'heating', 'electrical', 'ventilation', 'architecture']
49
+
50
+ You are a helpful assistant that understands BIM documents and engineering disciplines. Your answer should be in JSON format and only include the filename, a short description, and the engineering discipline the document belongs to, distinguishing between {[d.value for d in BimDiscipline]} based on the given document."
51
+
52
+ Analyze the provided document, which could be in either German or English. Extract the filename, its description, and infer the engineering discipline it belongs to. Document:
53
+ context="
54
+ """
55
+ # plain text
56
+ filepath = [doc.metadata for doc in docs][0]['source']
57
+ context = "".join(
58
+ [doc.page_content.replace('\n\n','').replace('..','') for doc in docs])
59
+
60
+ prompt = f'{prompt_template}{context}"\nFilepath:{filepath}'
61
+
62
+ #print(prompt)
63
+
64
+ # Create client
65
+ client = openai.OpenAI(
66
+ base_url="https://api.together.xyz/v1",
67
+ api_key=os.environ["TOGETHER_API_KEY"],
68
+ #api_key=userdata.get('TOGETHER_API_KEY'),
69
+ )
70
+
71
+ # Call the LLM with the JSON schema
72
+ chat_completion = client.chat.completions.create(
73
+ model="mistralai/Mixtral-8x7B-Instruct-v0.1",
74
+ messages=[
75
+ {
76
+ "role": "system",
77
+ "content": f"You are a helpful assistant that responsds in JSON format"
78
+ },
79
+ {
80
+ "role": "user",
81
+ "content": prompt
82
+ }
83
+ ]
84
+ )
85
+
86
+ return json.loads(chat_completion.choices[0].message.content)
87
+
88
+
89
+ if __name__ == "__main__":
90
+ parser = argparse.ArgumentParser(description="Generate metadata for a BIM document")
91
+ parser.add_argument("document", metavar="FILEPATH", type=str,
92
+ help="Path to the BIM document")
93
+
94
+ args = parser.parse_args()
95
+
96
+ if not os.path.exists(args.document) or not os.path.isfile(args.document):
97
+ print("File '{}' not found or not accessible.".format(args.document))
98
+ sys.exit(-1)
99
+
100
+ docs = ingest(args.document)
101
+ metadata = generate_metadata(docs)
102
+ print(metadata)