Spaces:
Runtime error
Runtime error
farhananis005
commited on
Commit
•
94d4a49
1
Parent(s):
74f3f5d
Update app.py
Browse files
app.py
CHANGED
@@ -1,36 +1,16 @@
|
|
1 |
-
# -*- coding: utf-8 -*-
|
2 |
-
"""Lawyer GPT
|
3 |
-
|
4 |
-
Automatically generated by Colab.
|
5 |
-
|
6 |
-
Original file is located at
|
7 |
-
https://colab.research.google.com/drive/1RPc_qH7s0_hsOswGpWRFaXbLT3eBIShJ
|
8 |
-
"""
|
9 |
-
|
10 |
-
!pip install langchain
|
11 |
-
!pip install langchain-openai
|
12 |
-
!pip install PyPDF2
|
13 |
-
!pip install pypdf
|
14 |
-
!pip install docx2txt
|
15 |
-
!pip install unstructured
|
16 |
-
!pip install gradio
|
17 |
-
!pip install faiss-cpu
|
18 |
-
!pip install openai
|
19 |
-
!pip install tiktoken
|
20 |
-
|
21 |
import os
|
22 |
import openai
|
23 |
|
24 |
os.environ["TOKENIZERS_PARALLELISM"] = "false"
|
25 |
-
os.environ["OPENAI_API_KEY"]
|
26 |
-
|
27 |
|
28 |
def save_docs(docs):
|
29 |
|
30 |
import shutil
|
31 |
import os
|
32 |
|
33 |
-
output_dir="/
|
34 |
|
35 |
if os.path.exists(output_dir):
|
36 |
shutil.rmtree(output_dir)
|
@@ -43,6 +23,7 @@ def save_docs(docs):
|
|
43 |
|
44 |
return "Successful!"
|
45 |
|
|
|
46 |
def process_docs():
|
47 |
|
48 |
from langchain.document_loaders import PyPDFLoader
|
@@ -55,19 +36,29 @@ def process_docs():
|
|
55 |
from langchain_openai import OpenAIEmbeddings
|
56 |
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
57 |
|
58 |
-
loader1 = DirectoryLoader(
|
|
|
|
|
59 |
document1 = loader1.load()
|
60 |
|
61 |
-
loader2 = DirectoryLoader(
|
|
|
|
|
62 |
document2 = loader2.load()
|
63 |
|
64 |
-
loader3 = DirectoryLoader(
|
|
|
|
|
65 |
document3 = loader3.load()
|
66 |
|
67 |
-
loader4 = DirectoryLoader(
|
|
|
|
|
68 |
document4 = loader4.load()
|
69 |
|
70 |
-
loader5 = DirectoryLoader(
|
|
|
|
|
71 |
document5 = loader5.load()
|
72 |
|
73 |
document1.extend(document2)
|
@@ -76,43 +67,45 @@ def process_docs():
|
|
76 |
document1.extend(document5)
|
77 |
|
78 |
text_splitter = RecursiveCharacterTextSplitter(
|
79 |
-
chunk_size=1000,
|
80 |
-
chunk_overlap=200,
|
81 |
-
length_function=len
|
82 |
)
|
83 |
|
84 |
docs = text_splitter.split_documents(document1)
|
85 |
embeddings = OpenAIEmbeddings()
|
86 |
|
87 |
docs_db = FAISS.from_documents(docs, embeddings)
|
88 |
-
docs_db.save_local("/
|
89 |
|
90 |
return "Successful!"
|
91 |
|
|
|
92 |
global agent
|
93 |
|
|
|
94 |
def create_agent():
|
95 |
|
96 |
from langchain_openai import ChatOpenAI
|
97 |
from langchain.chains.conversation.memory import ConversationSummaryBufferMemory
|
98 |
from langchain.chains import ConversationChain
|
|
|
99 |
global agent
|
100 |
|
101 |
-
llm = ChatOpenAI(model_name=
|
102 |
memory = ConversationSummaryBufferMemory(llm=llm, max_token_limit=1000)
|
103 |
agent = ConversationChain(llm=llm, memory=memory, verbose=True)
|
104 |
|
105 |
return "Successful!"
|
106 |
|
|
|
107 |
def formatted_response(docs, question, response, state):
|
108 |
|
109 |
formatted_output = response + "\n\nSources"
|
110 |
|
111 |
for i, doc in enumerate(docs):
|
112 |
-
source_info = doc.metadata.get(
|
113 |
-
page_info = doc.metadata.get(
|
114 |
|
115 |
-
doc_name = source_info.split(
|
116 |
|
117 |
if page_info is not None:
|
118 |
formatted_output += f"\n{doc_name}\tpage no {page_info}"
|
@@ -122,18 +115,22 @@ def formatted_response(docs, question, response, state):
|
|
122 |
state.append((question, formatted_output))
|
123 |
return state, state
|
124 |
|
|
|
125 |
def search_docs(prompt, question, state):
|
126 |
|
127 |
from langchain_openai import OpenAIEmbeddings
|
128 |
from langchain.vectorstores import FAISS
|
129 |
from langchain.callbacks import get_openai_callback
|
|
|
130 |
global agent
|
131 |
agent = agent
|
132 |
|
133 |
state = state or []
|
134 |
|
135 |
embeddings = OpenAIEmbeddings()
|
136 |
-
docs_db = FAISS.load_local(
|
|
|
|
|
137 |
docs = docs_db.similarity_search(question)
|
138 |
|
139 |
prompt += "\n\n"
|
@@ -147,6 +144,7 @@ def search_docs(prompt, question, state):
|
|
147 |
|
148 |
return formatted_response(docs, question, response, state)
|
149 |
|
|
|
150 |
import gradio as gr
|
151 |
|
152 |
css = """
|
@@ -164,45 +162,58 @@ with gr.Blocks(css=css) as demo:
|
|
164 |
gr.Markdown("## <center>Lawyer GPT: Your AI Legal Assistant</center>")
|
165 |
|
166 |
with gr.Tab("Lawyer GPT: Your AI Legal Assistant"):
|
167 |
-
|
168 |
|
169 |
-
|
170 |
-
|
171 |
|
172 |
-
|
173 |
-
|
174 |
-
|
175 |
|
176 |
-
|
177 |
-
|
178 |
|
179 |
-
|
180 |
-
|
181 |
|
182 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
183 |
|
184 |
-
|
185 |
-
|
186 |
|
187 |
-
|
188 |
|
189 |
-
|
190 |
-
|
191 |
|
192 |
-
|
193 |
-
|
194 |
|
195 |
-
|
196 |
|
197 |
#########################################################################################################
|
198 |
|
199 |
-
docs_upload_button.click(
|
|
|
|
|
200 |
docs_process_button.click(process_docs, inputs=None, outputs=docs_process_output)
|
201 |
create_agent_button.click(create_agent, inputs=None, outputs=create_agent_output)
|
202 |
|
203 |
-
docs_search_button.click(
|
|
|
|
|
|
|
|
|
204 |
|
205 |
#########################################################################################################
|
206 |
|
207 |
demo.queue()
|
208 |
-
demo.launch(debug=True, share=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import os
|
2 |
import openai
|
3 |
|
4 |
os.environ["TOKENIZERS_PARALLELISM"] = "false"
|
5 |
+
os.environ["OPENAI_API_KEY"]
|
6 |
+
|
7 |
|
8 |
def save_docs(docs):
|
9 |
|
10 |
import shutil
|
11 |
import os
|
12 |
|
13 |
+
output_dir = "/home/user/app/docs/"
|
14 |
|
15 |
if os.path.exists(output_dir):
|
16 |
shutil.rmtree(output_dir)
|
|
|
23 |
|
24 |
return "Successful!"
|
25 |
|
26 |
+
|
27 |
def process_docs():
|
28 |
|
29 |
from langchain.document_loaders import PyPDFLoader
|
|
|
36 |
from langchain_openai import OpenAIEmbeddings
|
37 |
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
38 |
|
39 |
+
loader1 = DirectoryLoader(
|
40 |
+
"/home/user/app/docs/", glob="./*.pdf", loader_cls=PyPDFLoader
|
41 |
+
)
|
42 |
document1 = loader1.load()
|
43 |
|
44 |
+
loader2 = DirectoryLoader(
|
45 |
+
"/home/user/app/docs/", glob="./*.txt", loader_cls=TextLoader
|
46 |
+
)
|
47 |
document2 = loader2.load()
|
48 |
|
49 |
+
loader3 = DirectoryLoader(
|
50 |
+
"/home/user/app/docs/", glob="./*.docx", loader_cls=Docx2txtLoader
|
51 |
+
)
|
52 |
document3 = loader3.load()
|
53 |
|
54 |
+
loader4 = DirectoryLoader(
|
55 |
+
"/home/user/app/docs/", glob="./*.csv", loader_cls=CSVLoader
|
56 |
+
)
|
57 |
document4 = loader4.load()
|
58 |
|
59 |
+
loader5 = DirectoryLoader(
|
60 |
+
"/home/user/app/docs/", glob="./*.xlsx", loader_cls=UnstructuredExcelLoader
|
61 |
+
)
|
62 |
document5 = loader5.load()
|
63 |
|
64 |
document1.extend(document2)
|
|
|
67 |
document1.extend(document5)
|
68 |
|
69 |
text_splitter = RecursiveCharacterTextSplitter(
|
70 |
+
chunk_size=1000, chunk_overlap=200, length_function=len
|
|
|
|
|
71 |
)
|
72 |
|
73 |
docs = text_splitter.split_documents(document1)
|
74 |
embeddings = OpenAIEmbeddings()
|
75 |
|
76 |
docs_db = FAISS.from_documents(docs, embeddings)
|
77 |
+
docs_db.save_local("/home/user/app/docs_db/")
|
78 |
|
79 |
return "Successful!"
|
80 |
|
81 |
+
|
82 |
global agent
|
83 |
|
84 |
+
|
85 |
def create_agent():
|
86 |
|
87 |
from langchain_openai import ChatOpenAI
|
88 |
from langchain.chains.conversation.memory import ConversationSummaryBufferMemory
|
89 |
from langchain.chains import ConversationChain
|
90 |
+
|
91 |
global agent
|
92 |
|
93 |
+
llm = ChatOpenAI(model_name="gpt-3.5-turbo-16k")
|
94 |
memory = ConversationSummaryBufferMemory(llm=llm, max_token_limit=1000)
|
95 |
agent = ConversationChain(llm=llm, memory=memory, verbose=True)
|
96 |
|
97 |
return "Successful!"
|
98 |
|
99 |
+
|
100 |
def formatted_response(docs, question, response, state):
|
101 |
|
102 |
formatted_output = response + "\n\nSources"
|
103 |
|
104 |
for i, doc in enumerate(docs):
|
105 |
+
source_info = doc.metadata.get("source", "Unknown source")
|
106 |
+
page_info = doc.metadata.get("page", None)
|
107 |
|
108 |
+
doc_name = source_info.split("/")[-1].strip()
|
109 |
|
110 |
if page_info is not None:
|
111 |
formatted_output += f"\n{doc_name}\tpage no {page_info}"
|
|
|
115 |
state.append((question, formatted_output))
|
116 |
return state, state
|
117 |
|
118 |
+
|
119 |
def search_docs(prompt, question, state):
|
120 |
|
121 |
from langchain_openai import OpenAIEmbeddings
|
122 |
from langchain.vectorstores import FAISS
|
123 |
from langchain.callbacks import get_openai_callback
|
124 |
+
|
125 |
global agent
|
126 |
agent = agent
|
127 |
|
128 |
state = state or []
|
129 |
|
130 |
embeddings = OpenAIEmbeddings()
|
131 |
+
docs_db = FAISS.load_local(
|
132 |
+
"/home/user/app/docs_db/", embeddings, allow_dangerous_deserialization=True
|
133 |
+
)
|
134 |
docs = docs_db.similarity_search(question)
|
135 |
|
136 |
prompt += "\n\n"
|
|
|
144 |
|
145 |
return formatted_response(docs, question, response, state)
|
146 |
|
147 |
+
|
148 |
import gradio as gr
|
149 |
|
150 |
css = """
|
|
|
162 |
gr.Markdown("## <center>Lawyer GPT: Your AI Legal Assistant</center>")
|
163 |
|
164 |
with gr.Tab("Lawyer GPT: Your AI Legal Assistant"):
|
165 |
+
with gr.Column(elem_classes="col"):
|
166 |
|
167 |
+
with gr.Tab("Upload and Process Documents"):
|
168 |
+
with gr.Column():
|
169 |
|
170 |
+
docs_upload_input = gr.Files(label="Upload File(s)")
|
171 |
+
docs_upload_button = gr.Button("Upload")
|
172 |
+
docs_upload_output = gr.Textbox(label="Output")
|
173 |
|
174 |
+
docs_process_button = gr.Button("Process")
|
175 |
+
docs_process_output = gr.Textbox(label="Output")
|
176 |
|
177 |
+
create_agent_button = gr.Button("Create Agent")
|
178 |
+
create_agent_output = gr.Textbox(label="Output")
|
179 |
|
180 |
+
gr.ClearButton(
|
181 |
+
[
|
182 |
+
docs_upload_input,
|
183 |
+
docs_upload_output,
|
184 |
+
docs_process_output,
|
185 |
+
create_agent_output,
|
186 |
+
]
|
187 |
+
)
|
188 |
|
189 |
+
with gr.Tab("Query Documents"):
|
190 |
+
with gr.Column():
|
191 |
|
192 |
+
docs_prompt_input = gr.Textbox(label="Custom Prompt")
|
193 |
|
194 |
+
docs_chatbot = gr.Chatbot(label="Chats")
|
195 |
+
docs_state = gr.State()
|
196 |
|
197 |
+
docs_search_input = gr.Textbox(label="Question")
|
198 |
+
docs_search_button = gr.Button("Search")
|
199 |
|
200 |
+
gr.ClearButton([docs_prompt_input, docs_search_input])
|
201 |
|
202 |
#########################################################################################################
|
203 |
|
204 |
+
docs_upload_button.click(
|
205 |
+
save_docs, inputs=docs_upload_input, outputs=docs_upload_output
|
206 |
+
)
|
207 |
docs_process_button.click(process_docs, inputs=None, outputs=docs_process_output)
|
208 |
create_agent_button.click(create_agent, inputs=None, outputs=create_agent_output)
|
209 |
|
210 |
+
docs_search_button.click(
|
211 |
+
search_docs,
|
212 |
+
inputs=[docs_prompt_input, docs_search_input, docs_state],
|
213 |
+
outputs=[docs_chatbot, docs_state],
|
214 |
+
)
|
215 |
|
216 |
#########################################################################################################
|
217 |
|
218 |
demo.queue()
|
219 |
+
demo.launch(debug=True, share=True)
|