Poonawala commited on
Commit
fd3b36c
·
verified ·
1 Parent(s): b76ac1b

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +158 -95
app.py CHANGED
@@ -1,27 +1,36 @@
1
  import gradio as gr
 
 
 
 
2
  from langchain_community.vectorstores import FAISS
3
  from langchain_community.document_loaders import PyPDFLoader
4
  from langchain.text_splitter import RecursiveCharacterTextSplitter
5
- from langchain_community.embeddings import HuggingFaceEmbeddings
6
  from langchain.chains import ConversationalRetrievalChain
7
- from langchain_community.llms import HuggingFaceEndpoint
 
8
  from langchain.chains import ConversationChain
9
  from langchain.memory import ConversationBufferMemory
10
- import os
11
-
12
- api_token = os.getenv("HF_TOKEN")
13
 
14
- # List of LLMs
15
- list_llm = ["meta-llama/Meta-Llama-3-8B-Instruct", "mistralai/Mistral-7B-Instruct-v0.2"]
16
  list_llm_simple = [os.path.basename(llm) for llm in list_llm]
17
 
18
- # Load and split PDF documents
19
  def load_doc(list_file_path):
 
 
 
20
  loaders = [PyPDFLoader(x) for x in list_file_path]
21
  pages = []
22
  for loader in loaders:
23
  pages.extend(loader.load())
24
- text_splitter = RecursiveCharacterTextSplitter(chunk_size=1024, chunk_overlap=64)
 
 
 
25
  doc_splits = text_splitter.split_documents(pages)
26
  return doc_splits
27
 
@@ -31,23 +40,24 @@ def create_db(splits):
31
  vectordb = FAISS.from_documents(splits, embeddings)
32
  return vectordb
33
 
34
- # Initialize LLM chain
35
- def initialize_llmchain(llm_model, temperature, max_tokens, top_k, vector_db):
 
36
  if llm_model == "meta-llama/Meta-Llama-3-8B-Instruct":
37
  llm = HuggingFaceEndpoint(
38
  repo_id=llm_model,
39
- huggingfacehub_api_token=api_token,
40
- temperature=temperature,
41
- max_new_tokens=max_tokens,
42
- top_k=top_k,
43
  )
44
  else:
45
  llm = HuggingFaceEndpoint(
46
- huggingfacehub_api_token=api_token,
47
- repo_id=llm_model,
48
- temperature=temperature,
49
- max_new_tokens=max_tokens,
50
- top_k=top_k,
51
  )
52
 
53
  memory = ConversationBufferMemory(
@@ -56,98 +66,151 @@ def initialize_llmchain(llm_model, temperature, max_tokens, top_k, vector_db):
56
  return_messages=True
57
  )
58
 
59
- retriever = vector_db.as_retriever()
60
  qa_chain = ConversationalRetrievalChain.from_llm(
61
  llm,
62
  retriever=retriever,
63
- chain_type="stuff",
64
  memory=memory,
65
  return_source_documents=True,
66
  verbose=False,
67
  )
68
  return qa_chain
69
 
70
- # Function to handle chatbot responses
71
- def respond(
72
- message,
73
- history: list[tuple[str, str]],
74
- system_message,
75
- max_tokens,
76
- temperature,
77
- top_p,
78
- vector_db,
79
- llm_model,
80
- ):
81
- # Initialize LLM chain if not already initialized
82
- if not hasattr(respond, 'qa_chain'):
83
- respond.qa_chain = initialize_llmchain(llm_model, temperature, max_tokens, top_p, vector_db)
84
-
85
- # Format chat history
 
 
 
 
86
  formatted_chat_history = []
87
- for user_message, bot_message in history:
88
  formatted_chat_history.append(f"User: {user_message}")
89
  formatted_chat_history.append(f"Assistant: {bot_message}")
90
- formatted_chat_history.append(f"User: {message}")
 
91
 
 
 
92
  # Generate response using QA chain
93
- response = respond.qa_chain.invoke({"question": message, "chat_history": formatted_chat_history})
94
  response_answer = response["answer"]
95
  if response_answer.find("Helpful Answer:") != -1:
96
  response_answer = response_answer.split("Helpful Answer:")[-1]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
97
 
98
- return response_answer
99
-
100
- # CSS for styling the interface
101
- css = """
102
- body {
103
- background-color: #06688E; /* Dark background */
104
- color: white; /* Text color for better visibility */
105
- }
106
- .gr-button {
107
- background-color: #42B3CE !important; /* White button color */
108
- color: black !important; /* Black text for contrast */
109
- border: none !important;
110
- padding: 8px 16px !important;
111
- border-radius: 5px !important;
112
- }
113
- .gr-button:hover {
114
- background-color: #e0e0e0 !important; /* Slightly lighter button on hover */
115
- }
116
- .gr-slider-container {
117
- color: white !important; /* Slider labels in white */
118
- }
119
- """
120
-
121
- # Initialize database and LLM chain
122
- def initialize_database_and_llm(list_file_obj, llm_option, max_tokens, temperature, top_p):
123
- list_file_path = [x.name for x in list_file_obj if x is not None]
124
- doc_splits = load_doc(list_file_path)
125
- vector_db = create_db(doc_splits)
126
- llm_name = list_llm[llm_option]
127
- return vector_db, llm_name
128
-
129
- # Gradio interface
130
- demo = gr.ChatInterface(
131
- respond,
132
- additional_inputs=[
133
- gr.Files(file_count="multiple", file_types=["pdf"], label="Upload PDF documents", visible=False),
134
- gr.Radio(list_llm_simple, label="Available LLMs", value=list_llm_simple, visible=False),
135
- gr.Slider(minimum=128, maximum=9192, value=4096, step=128, label="Max new tokens", visible=False),
136
- gr.Slider(minimum=0.01, maximum=1.0, value=0.5, step=0.1, label="Temperature", visible=False),
137
- gr.Slider(minimum=1, maximum=10, value=3, step=1, label="Top-k", visible=False),
138
- ],
139
- css=css,
140
- title="RAG PDF Chatbot",
141
- description="Query your PDF documents using a Retrieval Augmented Generation (RAG) chatbot.",
142
- )
143
-
144
- # Preprocessing events
145
- demo.preprocess(
146
- initialize_database_and_llm,
147
- inputs=["document", "llm_btn", "slider_maxtokens", "slider_temperature", "slider_topk"],
148
- outputs=["vector_db", "llm_model"],
149
- api_name="initialize",
150
- )
151
 
152
  if __name__ == "__main__":
153
- demo.launch(share=True)
 
1
  import gradio as gr
2
+ import os
3
+ api_token = os.getenv("HF_TOKEN")
4
+
5
+
6
  from langchain_community.vectorstores import FAISS
7
  from langchain_community.document_loaders import PyPDFLoader
8
  from langchain.text_splitter import RecursiveCharacterTextSplitter
9
+ from langchain_community.vectorstores import Chroma
10
  from langchain.chains import ConversationalRetrievalChain
11
+ from langchain_community.embeddings import HuggingFaceEmbeddings
12
+ from langchain_community.llms import HuggingFacePipeline
13
  from langchain.chains import ConversationChain
14
  from langchain.memory import ConversationBufferMemory
15
+ from langchain_community.llms import HuggingFaceEndpoint
16
+ import torch
 
17
 
18
+ list_llm = ["meta-llama/Meta-Llama-3-8B-Instruct", "mistralai/Mistral-7B-Instruct-v0.2"]
 
19
  list_llm_simple = [os.path.basename(llm) for llm in list_llm]
20
 
21
+ # Load and split PDF document
22
  def load_doc(list_file_path):
23
+ # Processing for one document only
24
+ # loader = PyPDFLoader(file_path)
25
+ # pages = loader.load()
26
  loaders = [PyPDFLoader(x) for x in list_file_path]
27
  pages = []
28
  for loader in loaders:
29
  pages.extend(loader.load())
30
+ text_splitter = RecursiveCharacterTextSplitter(
31
+ chunk_size = 1024,
32
+ chunk_overlap = 64
33
+ )
34
  doc_splits = text_splitter.split_documents(pages)
35
  return doc_splits
36
 
 
40
  vectordb = FAISS.from_documents(splits, embeddings)
41
  return vectordb
42
 
43
+
44
+ # Initialize langchain LLM chain
45
+ def initialize_llmchain(llm_model, temperature, max_tokens, top_k, vector_db, progress=gr.Progress()):
46
  if llm_model == "meta-llama/Meta-Llama-3-8B-Instruct":
47
  llm = HuggingFaceEndpoint(
48
  repo_id=llm_model,
49
+ huggingfacehub_api_token = api_token,
50
+ temperature = temperature,
51
+ max_new_tokens = max_tokens,
52
+ top_k = top_k,
53
  )
54
  else:
55
  llm = HuggingFaceEndpoint(
56
+ huggingfacehub_api_token = api_token,
57
+ repo_id=llm_model,
58
+ temperature = temperature,
59
+ max_new_tokens = max_tokens,
60
+ top_k = top_k,
61
  )
62
 
63
  memory = ConversationBufferMemory(
 
66
  return_messages=True
67
  )
68
 
69
+ retriever=vector_db.as_retriever()
70
  qa_chain = ConversationalRetrievalChain.from_llm(
71
  llm,
72
  retriever=retriever,
73
+ chain_type="stuff",
74
  memory=memory,
75
  return_source_documents=True,
76
  verbose=False,
77
  )
78
  return qa_chain
79
 
80
+ # Initialize database
81
+ def initialize_database(list_file_obj, progress=gr.Progress()):
82
+ # Create a list of documents (when valid)
83
+ list_file_path = [x.name for x in list_file_obj if x is not None]
84
+ # Load document and create splits
85
+ doc_splits = load_doc(list_file_path)
86
+ # Create or load vector database
87
+ vector_db = create_db(doc_splits)
88
+ return vector_db, "Database created!"
89
+
90
+ # Initialize LLM
91
+ def initialize_LLM(llm_option, llm_temperature, max_tokens, top_k, vector_db, progress=gr.Progress()):
92
+ # print("llm_option",llm_option)
93
+ llm_name = list_llm[llm_option]
94
+ print("llm_name: ",llm_name)
95
+ qa_chain = initialize_llmchain(llm_name, llm_temperature, max_tokens, top_k, vector_db, progress)
96
+ return qa_chain, "QA chain initialized. Chatbot is ready!"
97
+
98
+
99
+ def format_chat_history(message, chat_history):
100
  formatted_chat_history = []
101
+ for user_message, bot_message in chat_history:
102
  formatted_chat_history.append(f"User: {user_message}")
103
  formatted_chat_history.append(f"Assistant: {bot_message}")
104
+ return formatted_chat_history
105
+
106
 
107
+ def conversation(qa_chain, message, history):
108
+ formatted_chat_history = format_chat_history(message, history)
109
  # Generate response using QA chain
110
+ response = qa_chain.invoke({"question": message, "chat_history": formatted_chat_history})
111
  response_answer = response["answer"]
112
  if response_answer.find("Helpful Answer:") != -1:
113
  response_answer = response_answer.split("Helpful Answer:")[-1]
114
+ response_sources = response["source_documents"]
115
+ response_source1 = response_sources[0].page_content.strip()
116
+ response_source2 = response_sources[1].page_content.strip()
117
+ response_source3 = response_sources[2].page_content.strip()
118
+ # Langchain sources are zero-based
119
+ response_source1_page = response_sources[0].metadata["page"] + 1
120
+ response_source2_page = response_sources[1].metadata["page"] + 1
121
+ response_source3_page = response_sources[2].metadata["page"] + 1
122
+ # Append user message and response to chat history
123
+ new_history = history + [(message, response_answer)]
124
+ return qa_chain, gr.update(value=""), new_history, response_source1, response_source1_page, response_source2, response_source2_page, response_source3, response_source3_page
125
+
126
+
127
+ def upload_file(file_obj):
128
+ list_file_path = []
129
+ for idx, file in enumerate(file_obj):
130
+ file_path = file_obj.name
131
+ list_file_path.append(file_path)
132
+ return list_file_path
133
+
134
+
135
+ def demo():
136
+ # with gr.Blocks(theme=gr.themes.Default(primary_hue="sky")) as demo:
137
+ with gr.Blocks(theme=gr.themes.Default(primary_hue="red", secondary_hue="pink", neutral_hue = "sky")) as demo:
138
+ vector_db = gr.State()
139
+ qa_chain = gr.State()
140
+ gr.HTML("<center><h1>RAG PDF chatbot</h1><center>")
141
+ gr.Markdown("""<b>Query your PDF documents!</b> This AI agent is designed to perform retrieval augmented generation (RAG) on PDF documents. The app is hosted on Hugging Face Hub for the sole purpose of demonstration. \
142
+ <b>Please do not upload confidential documents.</b>
143
+ """)
144
+ with gr.Row():
145
+ with gr.Column(scale = 86):
146
+ gr.Markdown("<b>Step 1 - Upload PDF documents and Initialize RAG pipeline</b>")
147
+ with gr.Row():
148
+ document = gr.Files(height=300, file_count="multiple", file_types=["pdf"], interactive=True, label="Upload PDF documents")
149
+ with gr.Row():
150
+ db_btn = gr.Button("Create vector database")
151
+ with gr.Row():
152
+ db_progress = gr.Textbox(value="Not initialized", show_label=False) # label="Vector database status",
153
+ gr.Markdown("<style>body { font-size: 16px; }</style><b>Select Large Language Model (LLM) and input parameters</b>")
154
+ with gr.Row():
155
+ llm_btn = gr.Radio(list_llm_simple, label="Available LLMs", value = list_llm_simple[0], type="index") # info="Select LLM", show_label=False
156
+ with gr.Row():
157
+ with gr.Accordion("LLM input parameters", open=False):
158
+ with gr.Row():
159
+ slider_temperature = gr.Slider(minimum = 0.01, maximum = 1.0, value=0.5, step=0.1, label="Temperature", info="Controls randomness in token generation", interactive=True)
160
+ with gr.Row():
161
+ slider_maxtokens = gr.Slider(minimum = 128, maximum = 9192, value=4096, step=128, label="Max New Tokens", info="Maximum number of tokens to be generated",interactive=True)
162
+ with gr.Row():
163
+ slider_topk = gr.Slider(minimum = 1, maximum = 10, value=3, step=1, label="top-k", info="Number of tokens to select the next token from", interactive=True)
164
+ with gr.Row():
165
+ qachain_btn = gr.Button("Initialize Question Answering Chatbot")
166
+ with gr.Row():
167
+ llm_progress = gr.Textbox(value="Not initialized", show_label=False) # label="Chatbot status",
168
+
169
+ with gr.Column(scale = 200):
170
+ gr.Markdown("<b>Step 2 - Chat with your Document</b>")
171
+ chatbot = gr.Chatbot(height=505)
172
+ with gr.Accordion("Relevent context from the source document", open=False):
173
+ with gr.Row():
174
+ doc_source1 = gr.Textbox(label="Reference 1", lines=2, container=True, scale=20)
175
+ source1_page = gr.Number(label="Page", scale=1)
176
+ with gr.Row():
177
+ doc_source2 = gr.Textbox(label="Reference 2", lines=2, container=True, scale=20)
178
+ source2_page = gr.Number(label="Page", scale=1)
179
+ with gr.Row():
180
+ doc_source3 = gr.Textbox(label="Reference 3", lines=2, container=True, scale=20)
181
+ source3_page = gr.Number(label="Page", scale=1)
182
+ with gr.Row():
183
+ msg = gr.Textbox(placeholder="Ask a question", container=True)
184
+ with gr.Row():
185
+ submit_btn = gr.Button("Submit")
186
+ clear_btn = gr.ClearButton([msg, chatbot], value="Clear")
187
+
188
+ # Preprocessing events
189
+ db_btn.click(initialize_database, \
190
+ inputs=[document], \
191
+ outputs=[vector_db, db_progress])
192
+ qachain_btn.click(initialize_LLM, \
193
+ inputs=[llm_btn, slider_temperature, slider_maxtokens, slider_topk, vector_db], \
194
+ outputs=[qa_chain, llm_progress]).then(lambda:[None,"",0,"",0,"",0], \
195
+ inputs=None, \
196
+ outputs=[chatbot, doc_source1, source1_page, doc_source2, source2_page, doc_source3, source3_page], \
197
+ queue=False)
198
+
199
+ # Chatbot events
200
+ msg.submit(conversation, \
201
+ inputs=[qa_chain, msg, chatbot], \
202
+ outputs=[qa_chain, msg, chatbot, doc_source1, source1_page, doc_source2, source2_page, doc_source3, source3_page], \
203
+ queue=False)
204
+ submit_btn.click(conversation, \
205
+ inputs=[qa_chain, msg, chatbot], \
206
+ outputs=[qa_chain, msg, chatbot, doc_source1, source1_page, doc_source2, source2_page, doc_source3, source3_page], \
207
+ queue=False)
208
+ clear_btn.click(lambda:[None,"",0,"",0,"",0], \
209
+ inputs=None, \
210
+ outputs=[chatbot, doc_source1, source1_page, doc_source2, source2_page, doc_source3, source3_page], \
211
+ queue=False)
212
+ demo.queue().launch(debug=True)
213
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
214
 
215
  if __name__ == "__main__":
216
+ demo()