Spaces:
Running
Running
Shivanand Roy
commited on
Commit
•
bc69d30
1
Parent(s):
f98d774
Added application file
Browse files- Dockerfile +11 -0
- app.py +145 -0
- assets/ChatPDF.jpg +0 -0
- assets/ChatPDFAvatar.jpg +0 -0
- assets/ChatPDFAvatar.png +0 -0
- assets/ChatPDFLogo.png +0 -0
- assets/ChatPDFLogoV2.png +0 -0
- assets/avatar.jpg +0 -0
- chainlit.md +19 -0
Dockerfile
ADDED
@@ -0,0 +1,11 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
FROM python:3.9
|
2 |
+
RUN useradd -m -u 1000 user
|
3 |
+
USER user
|
4 |
+
ENV HOME=/home/user \
|
5 |
+
PATH=/home/user/.local/bin:$PATH
|
6 |
+
WORKDIR $HOME/app
|
7 |
+
COPY --chown=user . $HOME/app
|
8 |
+
COPY ./requirements.txt ~/app/requirements.txt
|
9 |
+
RUN pip install -r requirements.txt
|
10 |
+
COPY . .
|
11 |
+
CMD ["chainlit", "run", "app.py", "--port", "7860"]
|
app.py
ADDED
@@ -0,0 +1,145 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from langchain.embeddings.openai import OpenAIEmbeddings
|
2 |
+
from langchain.text_splitter import RecursiveCharacterTextSplitter, CharacterTextSplitter
|
3 |
+
from langchain.vectorstores import Chroma
|
4 |
+
from langchain.chains import RetrievalQAWithSourcesChain
|
5 |
+
from langchain.memory import ConversationBufferWindowMemory
|
6 |
+
from langchain.chains import ConversationalRetrievalChain
|
7 |
+
from langchain.chat_models import ChatOpenAI
|
8 |
+
from langchain.prompts.chat import (
|
9 |
+
ChatPromptTemplate,
|
10 |
+
SystemMessagePromptTemplate,
|
11 |
+
HumanMessagePromptTemplate,
|
12 |
+
)
|
13 |
+
from langchain.document_loaders import PyPDFLoader
|
14 |
+
import os
|
15 |
+
import chainlit as cl
|
16 |
+
from langchain.prompts import PromptTemplate
|
17 |
+
|
18 |
+
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=100)
|
19 |
+
|
20 |
+
system_template = """Use the following pieces of context to answer the users question.
|
21 |
+
If you don't know the answer, just say that you don't know, don't try to make up an answer.
|
22 |
+
ALWAYS return a "SOURCES" part in your answer.
|
23 |
+
The "SOURCES" part should be a reference to the source of the document from which you got your answer.
|
24 |
+
|
25 |
+
Example of your response should be:
|
26 |
+
|
27 |
+
```
|
28 |
+
The answer is foo
|
29 |
+
SOURCES: xyz
|
30 |
+
```
|
31 |
+
|
32 |
+
Begin!
|
33 |
+
----------------
|
34 |
+
{summaries}"""
|
35 |
+
messages = [
|
36 |
+
SystemMessagePromptTemplate.from_template(system_template),
|
37 |
+
HumanMessagePromptTemplate.from_template("{question}"),
|
38 |
+
]
|
39 |
+
prompt = ChatPromptTemplate.from_messages(messages)
|
40 |
+
chain_type_kwargs = {"prompt": prompt}
|
41 |
+
|
42 |
+
@cl.on_chat_start
|
43 |
+
async def start():
|
44 |
+
await cl.Avatar(
|
45 |
+
name="ChatPDF",
|
46 |
+
url="https://avatars.githubusercontent.com/u/128686189?s=400&u=a1d1553023f8ea0921fba0debbe92a8c5f840dd9&v=4",
|
47 |
+
# path = r'assets/ChatPDFAvatar.jpg'
|
48 |
+
).send()
|
49 |
+
|
50 |
+
|
51 |
+
@cl.langchain_factory(use_async=True)
|
52 |
+
async def init():
|
53 |
+
files = None
|
54 |
+
|
55 |
+
# Wait for the user to upload a file
|
56 |
+
while files == None:
|
57 |
+
files = await cl.AskFileMessage(
|
58 |
+
content="Hey, Welcome to ChatPDF!\n\nChatPDF is a smart, user-friendly tool that integrates state-of-the-art AI models with text extraction and embedding capabilities to create a unique, conversational interaction with your PDF documents.\n\nSimply upload your PDF, ask your questions, and ChatPDF will deliver the most relevant answers directly from your document.\n\nPlease upload a PDF file to begin!", accept=["application/pdf"]
|
59 |
+
).send()
|
60 |
+
|
61 |
+
file = files[0]
|
62 |
+
|
63 |
+
msg = cl.Message(content=f'''Processing "{file.name}"...''')
|
64 |
+
await msg.send()
|
65 |
+
|
66 |
+
#
|
67 |
+
|
68 |
+
with open(os.path.join(file.name), "wb") as f:
|
69 |
+
f.write(file.content)
|
70 |
+
|
71 |
+
print(file.name)
|
72 |
+
|
73 |
+
loader = PyPDFLoader(file.name)
|
74 |
+
pages = loader.load_and_split()
|
75 |
+
|
76 |
+
# add page split info
|
77 |
+
# Initialize a dictionary to keep track of duplicate page numbers
|
78 |
+
page_counts = {}
|
79 |
+
|
80 |
+
for document in pages:
|
81 |
+
page_number = document.metadata['page']
|
82 |
+
|
83 |
+
# If this is the first occurrence of this page number, initialize its count to 1
|
84 |
+
# Otherwise, increment the count for this page number
|
85 |
+
page_counts[page_number] = page_counts.get(page_number, 0) + 1
|
86 |
+
|
87 |
+
# Create the page split info string
|
88 |
+
page_split_info = f"Page-{page_number+1}.{page_counts[page_number]}"
|
89 |
+
|
90 |
+
# Add the page split info to the document's metadata
|
91 |
+
document.metadata['page_split_info'] = page_split_info
|
92 |
+
|
93 |
+
|
94 |
+
|
95 |
+
# Create a Chroma vector store
|
96 |
+
embeddings = OpenAIEmbeddings()
|
97 |
+
docsearch = await cl.make_async(Chroma.from_documents)(
|
98 |
+
pages, embeddings
|
99 |
+
)
|
100 |
+
|
101 |
+
# define memory
|
102 |
+
memory = ConversationBufferWindowMemory(
|
103 |
+
k=5,
|
104 |
+
memory_key='chat_history',
|
105 |
+
return_messages=True,
|
106 |
+
output_key='answer'
|
107 |
+
)
|
108 |
+
|
109 |
+
# Create a chain that uses the Chroma vector store
|
110 |
+
chain = ConversationalRetrievalChain.from_llm(
|
111 |
+
ChatOpenAI(temperature=0, model="gpt-3.5-turbo-16k", streaming=True),
|
112 |
+
chain_type="stuff",
|
113 |
+
retriever=docsearch.as_retriever(search_kwargs={'k':5}),
|
114 |
+
memory=memory,
|
115 |
+
return_source_documents=True,
|
116 |
+
)
|
117 |
+
|
118 |
+
# Save the metadata and texts in the user session
|
119 |
+
# cl.user_session.set("metadatas", metadatas)
|
120 |
+
cl.user_session.set("texts", pages)
|
121 |
+
|
122 |
+
# Let the user know that the system is ready
|
123 |
+
await msg.update(content=f''' "{file.name}" processed. You can now ask questions!''')
|
124 |
+
|
125 |
+
|
126 |
+
return chain
|
127 |
+
|
128 |
+
|
129 |
+
@cl.langchain_postprocess
|
130 |
+
async def process_response(res):
|
131 |
+
answer = res["answer"]
|
132 |
+
source_documents = res['source_documents']
|
133 |
+
content = [source_documents[i].page_content for i in range(len(source_documents))]
|
134 |
+
name = [source_documents[i].metadata['page_split_info'] for i in range(len(source_documents))]
|
135 |
+
source_elements = [
|
136 |
+
cl.Text(content=content[i], name=name[i]) for i in range(len(source_documents))
|
137 |
+
]
|
138 |
+
|
139 |
+
if source_documents:
|
140 |
+
answer += f"\n\nSources: {', '.join([source_documents[i].metadata['page_split_info'] for i in range(len(source_documents))])}"
|
141 |
+
else:
|
142 |
+
answer += "\n\nNo sources found"
|
143 |
+
|
144 |
+
await cl.Message(content=answer, elements=source_elements).send()
|
145 |
+
# await cl.Message(content=answer).send()
|
assets/ChatPDF.jpg
ADDED
![]() |
assets/ChatPDFAvatar.jpg
ADDED
![]() |
assets/ChatPDFAvatar.png
ADDED
![]() |
assets/ChatPDFLogo.png
ADDED
![]() |
assets/ChatPDFLogoV2.png
ADDED
![]() |
assets/avatar.jpg
ADDED
![]() |
chainlit.md
ADDED
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# ChatPDF - Conversation Style Question Answering with PDFs
|
2 |
+
|
3 |
+
ChatPDF is an application designed to provide super fast, conversation-style question answering from any PDF documents.
|
4 |
+
|
5 |
+
![](assets/ChatPDF.jpg)
|
6 |
+
|
7 |
+
ChatPDF is a smart, user-friendly tool that integrates state-of-the-art AI models with text extraction and embedding capabilities to create a unique, conversational interaction with your PDF documents.
|
8 |
+
|
9 |
+
Simply upload your PDF, ask your questions, and ChatPDF will deliver the most relevant answers directly from your document.
|
10 |
+
|
11 |
+
|
12 |
+
## Features
|
13 |
+
|
14 |
+
1. PDF text extraction: ChatPDF accepts PDF files, from which it extracts the text in an intelligent manner.
|
15 |
+
2. Document chunking: The extracted text is split into manageable chunks, enabling efficient processing.
|
16 |
+
3. Text embedding: The chunks are embedded using sophisticated natural language processing techniques and stored in a vector database.
|
17 |
+
4. Fast querying: Upon user query, the application swiftly identifies the most relevant document chunks from the vector database.
|
18 |
+
5. AI-powered answers: The selected chunks are then passed to a large language model (LLM), which generates detailed, coherent responses.
|
19 |
+
|