Create app.py

app.py

@@ -0,0 +1,142 @@
+import streamlit as st 
+import databutton as db
+
+from langchain.llms import OpenAI
+from langchain.chains import RetrievalQA
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+from langchain.document_loaders import YoutubeLoader
+from langchain.docstore.document import Document
+from langchain.vectorstores import FAISS
+from langchain.embeddings import OpenAIEmbeddings
+
+
+def text_custom(font_size, text):
+    '''
+    font_size := ['b', 'm', 's']
+    '''
+    result=f'<p class="{font_size}-font">{text}</p>'
+    return result
+
+
+def main():
+
+    st.set_page_config(
+        page_title="Welcome to Question Answering on YouTube video page",
+        layout="wide",  # {wide, centered}
+    )
+    # reference
+    ## https://discuss.streamlit.io/t/change-input-text-font-size/29959/4
+    ## https://discuss.streamlit.io/t/change-font-size-in-st-write/7606/2
+    st.markdown("""<style>.b-font {font-size:25px !important;}</style>""", unsafe_allow_html=True)    
+    st.markdown("""<style>.m-font {font-size:20px !important;}</style>""" , unsafe_allow_html=True)    
+    st.markdown("""<style>.s-font {font-size:15px !important;}</style>""" , unsafe_allow_html=True)    
+    tabs_font_css = """<style>div[class*="stTextInput"] label {font-size: 15px;color: black;}</style>"""
+    st.write(tabs_font_css, unsafe_allow_html=True)
+
+    st.title("YouTube QnA")
+    t = "Watch all Youtube videos... Sometimes it's hard, right? Just throw us a URL and ask. I'll answer anything. 😋"
+    st.markdown(text_custom('m', t), unsafe_allow_html=True)
+    st.info('Note: This youtube video itself should have transcript', icon="ℹ️")
+
+    api_key = st.text_input(
+        "Enter Open AI Key.",
+        placeholder = "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",
+        type="password"
+    )
+
+    user_in_url = st.text_input(
+        "Please enter Youtube URL.",
+        value = "https://www.youtube.com/watch?v=o8NPllzkFhE",
+    )
+
+    if user_in_url:
+        width = 40
+        side = max((100 - width) / 2, 0.01)
+        _, container, _ = st.columns([side, width, side])
+        container.video(data=user_in_url)
+
+    user_question = st.text_input(
+            "Please enter your questions in the video.",
+            placeholder = "What is the Linux and Why it is created?"
+    )
+
+    user_in_lang = st.text_input(
+        "Tell us what language the Youtube video is in (For example.. enter 'en' for English or 'hi' for Hindi).",
+        value = "en",
+    )    
+
+    with st.sidebar:
+        embeddeing_model = st.selectbox(
+            label='Embedding Model',
+            options=['text-embedding-ada-002']
+        )
+
+        llm_model = st.selectbox(
+            label='LLM Model',
+            options=["text-davinci-003",
+                     "text-curie-001",
+                     "text-babbage-001",
+                     "text-ada-001"]
+        )
+
+        chain = st.radio(
+            label='Chain type',
+            options=['stuff',
+                    'map_reduce',
+                    'refine']
+        )
+
+        temperature = st.slider(
+            "Temperature",
+            0.0, 1.0, 0.7,
+        )
+
+
+    if st.button("Hi GPT! Answer the question right now."):
+        API=api_key
+        if not API:
+            st.warning("Enter your OPENAI API-KEY. If you don't have one Get your OpenAI API key from [here](https://platform.openai.com/account/api-keys).")
+
+
+        # 1. get text data from external source(Youtube video transcription)
+        # 참고: https://python.langchain.com/en/latest/modules/indexes/document_loaders.html
+        documents = YoutubeLoader.from_youtube_url(user_in_url, language=user_in_lang).load()
+
+
+        # 2. text preprocessing(Chunking)
+        text_splitter = RecursiveCharacterTextSplitter(
+            chunk_size=800,
+            separators=['\n\n', '\n', '.', '!', '?', ',', ' ', ''],
+            chunk_overlap=200
+        )
+        docs=text_splitter.split_text(documents[0].page_content)
+        new_docs = [Document(page_content=chunk) for chunk in docs]
+        
+
+        # 3. define embedding model & provider
+        embeddings = OpenAIEmbeddings(openai_api_key=API, model=embeddeing_model)
+        
+        # 4. create embedding vectorstore(Vector DB) to use as the index
+        db = FAISS.from_documents(new_docs, embeddings)
+
+        # 5. Make chin for `question-answering` task with an information retriever
+        retriever = db.as_retriever()
+
+        qa = RetrievalQA.from_chain_type(
+            llm=OpenAI(openai_api_key=API, 
+                       model=llm_model, 
+                       temperature=temperature, 
+                       verbose=True),
+            chain_type=chain, 
+            retriever=retriever, 
+            return_source_documents=True,
+            verbose=True)
+         
+        with st.spinner("Running to answer your question .."):
+            query = user_question
+            result = qa({"query": user_question})
+            st.success(result['result'])
+
+
+if __name__=='__main__':
+    main()