Upload 2 files

app.py

@@ -0,0 +1,208 @@
+# -*- coding: utf-8 -*-
+"""Lawyer GPT
+
+Automatically generated by Colab.
+
+Original file is located at
+    https://colab.research.google.com/drive/1RPc_qH7s0_hsOswGpWRFaXbLT3eBIShJ
+"""
+
+!pip install langchain
+!pip install langchain-openai
+!pip install PyPDF2
+!pip install pypdf
+!pip install docx2txt
+!pip install unstructured
+!pip install gradio
+!pip install faiss-cpu
+!pip install openai
+!pip install tiktoken
+
+import os
+import openai
+
+os.environ["TOKENIZERS_PARALLELISM"] = "false"
+os.environ["OPENAI_API_KEY"] = "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"
+openai.api_key = "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"
+
+def save_docs(docs):
+
+    import shutil
+    import os
+
+    output_dir="/content/docs/"
+
+    if os.path.exists(output_dir):
+        shutil.rmtree(output_dir)
+
+    if not os.path.exists(output_dir):
+        os.makedirs(output_dir)
+
+    for doc in docs:
+        shutil.copy(doc.name, output_dir)
+
+    return "Successful!"
+
+def process_docs():
+
+    from langchain.document_loaders import PyPDFLoader
+    from langchain.document_loaders import DirectoryLoader
+    from langchain.document_loaders import TextLoader
+    from langchain.document_loaders import Docx2txtLoader
+    from langchain.document_loaders.csv_loader import CSVLoader
+    from langchain.document_loaders import UnstructuredExcelLoader
+    from langchain.vectorstores import FAISS
+    from langchain_openai import OpenAIEmbeddings
+    from langchain.text_splitter import RecursiveCharacterTextSplitter
+
+    loader1 = DirectoryLoader('/content/docs/', glob="./*.pdf", loader_cls=PyPDFLoader)
+    document1 = loader1.load()
+
+    loader2 = DirectoryLoader('/content/docs/', glob="./*.txt", loader_cls=TextLoader)
+    document2 = loader2.load()
+
+    loader3 = DirectoryLoader('/content/docs/', glob="./*.docx", loader_cls=Docx2txtLoader)
+    document3 = loader3.load()
+
+    loader4 = DirectoryLoader('/content/docs/', glob="./*.csv", loader_cls=CSVLoader)
+    document4 = loader4.load()
+
+    loader5 = DirectoryLoader('/content/docs/', glob="./*.xlsx", loader_cls=UnstructuredExcelLoader)
+    document5 = loader5.load()
+
+    document1.extend(document2)
+    document1.extend(document3)
+    document1.extend(document4)
+    document1.extend(document5)
+
+    text_splitter = RecursiveCharacterTextSplitter(
+        chunk_size=1000,
+        chunk_overlap=200,
+        length_function=len
+    )
+
+    docs = text_splitter.split_documents(document1)
+    embeddings = OpenAIEmbeddings()
+
+    docs_db = FAISS.from_documents(docs, embeddings)
+    docs_db.save_local("/content/docs_db/")
+
+    return "Successful!"
+
+global agent
+
+def create_agent():
+
+    from langchain_openai import ChatOpenAI
+    from langchain.chains.conversation.memory import ConversationSummaryBufferMemory
+    from langchain.chains import ConversationChain
+    global agent
+
+    llm = ChatOpenAI(model_name='gpt-3.5-turbo-16k')
+    memory = ConversationSummaryBufferMemory(llm=llm, max_token_limit=1000)
+    agent = ConversationChain(llm=llm, memory=memory, verbose=True)
+
+    return "Successful!"
+
+def formatted_response(docs, question, response, state):
+
+    formatted_output = response + "\n\nSources"
+
+    for i, doc in enumerate(docs):
+        source_info = doc.metadata.get('source', 'Unknown source')
+        page_info = doc.metadata.get('page', None)
+
+        doc_name = source_info.split('/')[-1].strip()
+
+        if page_info is not None:
+            formatted_output += f"\n{doc_name}\tpage no {page_info}"
+        else:
+            formatted_output += f"\n{doc_name}"
+
+    state.append((question, formatted_output))
+    return state, state
+
+def search_docs(prompt, question, state):
+
+    from langchain_openai import OpenAIEmbeddings
+    from langchain.vectorstores import FAISS
+    from langchain.callbacks import get_openai_callback
+    global agent
+    agent = agent
+
+    state = state or []
+
+    embeddings = OpenAIEmbeddings()
+    docs_db = FAISS.load_local("/content/docs_db/", embeddings, allow_dangerous_deserialization = True)
+    docs = docs_db.similarity_search(question)
+
+    prompt += "\n\n"
+    prompt += question
+    prompt += "\n\n"
+    prompt += str(docs)
+
+    with get_openai_callback() as cb:
+        response = agent.predict(input=prompt)
+        print(cb)
+
+    return formatted_response(docs, question, response, state)
+
+import gradio as gr
+
+css = """
+.col{
+    max-width: 75%;
+    margin: 0 auto;
+    display: flex;
+    flex-direction: column;
+    justify-content: center;
+    align-items: center;
+}
+"""
+
+with gr.Blocks(css=css) as demo:
+    gr.Markdown("## <center>Lawyer GPT: Your AI Legal Assistant</center>")
+
+    with gr.Tab("Lawyer GPT: Your AI Legal Assistant"):
+      with gr.Column(elem_classes="col"):
+
+        with gr.Tab("Upload and Process Documents"):
+          with gr.Column():
+
+            docs_upload_input = gr.Files(label="Upload File(s)")
+            docs_upload_button = gr.Button("Upload")
+            docs_upload_output = gr.Textbox(label="Output")
+
+            docs_process_button = gr.Button("Process")
+            docs_process_output = gr.Textbox(label="Output")
+
+            create_agent_button = gr.Button("Create Agent")
+            create_agent_output = gr.Textbox(label="Output")
+
+            gr.ClearButton([docs_upload_input, docs_upload_output, docs_process_output, create_agent_output])
+
+        with gr.Tab("Query Documents"):
+          with gr.Column():
+
+            docs_prompt_input = gr.Textbox(label="Custom Prompt")
+
+            docs_chatbot = gr.Chatbot(label="Chats")
+            docs_state = gr.State()
+
+            docs_search_input = gr.Textbox(label="Question")
+            docs_search_button = gr.Button("Search")
+
+            gr.ClearButton([docs_prompt_input, docs_search_input])
+
+    #########################################################################################################
+
+    docs_upload_button.click(save_docs, inputs=docs_upload_input, outputs=docs_upload_output)
+    docs_process_button.click(process_docs, inputs=None, outputs=docs_process_output)
+    create_agent_button.click(create_agent, inputs=None, outputs=create_agent_output)
+
+    docs_search_button.click(search_docs, inputs=[docs_prompt_input, docs_search_input, docs_state], outputs=[docs_chatbot, docs_state])
+
+    #########################################################################################################
+
+demo.queue()
+demo.launch(debug=True, share=True)

requirements.txt

@@ -0,0 +1,10 @@
+langchain
+langchain-openai
+PyPDF2
+pypdf
+docx2txt
+unstructured
+gradio
+faiss-cpu
+openai
+tiktoken