lllm2 / src /app.py
lotran12's picture
Upload 9 files
b8228d1 verified
Raw History Blame Contribute Delete
5.47 kB
import os
import gradio as gr
from base.llm_model import get_llm
from rag.rag_module import Offline_RAG
from rag.file_loader import DocumentLoader
from rag.vector_store import VectorDB
from langchain_google_genai import GoogleGenerativeAIEmbeddings
os.environ["TOKENIZERS_PARALLELISM"] = "false"
llm = get_llm(
model_name="gemini-1.5-flash-002",
temperature=0.9,
max_output_tokens=8192,
)
class ChatInterface:
def __init__(self):
self.rag_chain = None
self.chat_chain = None
self.vector_store = VectorDB(embedding=GoogleGenerativeAIEmbeddings(model="models/text-embedding-004"))
self.chat_history = []
def process_files(self, files, url_text):
documents = []
if files:
doc_loader = DocumentLoader(file_type='pdf')
pdf_paths = [file.name for file in files]
documents.extend(doc_loader.load(pdf_paths))
if url_text:
html_loader = DocumentLoader(file_type='html')
documents.extend(html_loader.load([url_text]))
if documents:
self.vector_store.update_db(documents)
return [{'role': 'assistant', 'content': 'Documents processed successfully!'}]
return [{'role': 'assistant', 'content': 'No documents provided.'}]
def get_response(self, user_input, mode):
if not user_input.strip():
return [{'role': 'assistant', 'content': 'Chat history cleared'}]
if mode == "RAG" and self.vector_store.db is None:
self.chat_history.append({"role": "user", "content": user_input})
self.chat_history.append({"role": "assistant", "content": "Please upload documents first for RAG mode."})
return [{'role': 'assistant', 'content': 'Chat history cleared'}]
try:
if mode == "RAG":
chain = self.rag_chain or Offline_RAG(llm, type='rag').get_chain(self.vector_store.get_retriever())
else:
chain = self.chat_chain or Offline_RAG(llm, type='chat').get_chain()
history_text = "\n".join(
f'User: {m["content"]}' if m["role"]=="user" else f'Assistant: {m["content"]}'
for m in self.chat_history
)
response = chain.invoke({
"chat_history": history_text,
"input": user_input
})
self.chat_history.append({"role": "user", "content": user_input})
self.chat_history.append({"role": "assistant", "content": response})
except Exception as e:
self.chat_history.append({"role": "user", "content": user_input})
self.chat_history.append({"role": "assistant", "content": f"Error: {str(e)}"})
return [{'role': 'assistant', 'content': 'Chat history cleared'}]
def clear_history(self):
self.chat_history = []
return []
def clear_vector_store(self):
self.vector_store.db = None
return [{'role': 'assistant', 'content': 'Vector store cleared successfully!'}]
def main():
chat_interface = ChatInterface()
with gr.Blocks() as demo:
gr.Markdown(
"""
# Document Chat Interface
Upload documents or provide URLs to chat with their content using RAG or standard chatbot mode.
"""
)
with gr.Row():
with gr.Column(scale=1):
with gr.Tab('PDF Files'):
pdf_docs = gr.File(label="Upload PDF Files", file_count="multiple")
with gr.Tab('URLs'):
url_text = gr.Textbox(label="Or Enter URL", placeholder="https://example.com", lines=1)
mode_switch = gr.Radio(["chatbot", "RAG"], label="Select Mode", value="chatbot")
submit_button = gr.Button("Process Documents")
with gr.Column(scale=4):
chat_interface_md = gr.Markdown(
"""
### Instructions
1. Upload PDF files or enter URLs
2. Click 'Process Documents' to analyze the content
3. Select mode (RAG or chatbot)
4. Start chatting!
"""
)
chat_history = gr.Chatbot(label="Chat History", height=400, type="messages")
with gr.Row():
user_input = gr.Textbox(label="Your Question", placeholder="Ask something...", lines=2, scale=5)
submit_chat = gr.Button("Send", scale=1)
clear_button = gr.Button("Clear Chat")
clear_vector_button = gr.Button("Clear Vector Store")
submit_button.click(
chat_interface.process_files,
inputs=[pdf_docs, url_text],
outputs=chat_history
)
user_input.submit(
chat_interface.get_response,
inputs=[user_input, mode_switch],
outputs=chat_history
)
submit_chat.click(
chat_interface.get_response,
inputs=[user_input, mode_switch],
outputs=chat_history
)
clear_button.click(
chat_interface.clear_history,
outputs=chat_history
)
clear_vector_button.click(
chat_interface.clear_vector_store,
outputs=chat_history
)
demo.launch()
if __name__ == "__main__":
main()