TUTORIAL60

tutorial2/single_url.py

@@ -1,7 +1,7 @@
 from langchain.document_loaders import youtube
 import io
 
-loader=youtube.YoutubeLoader.from_youtube_url("https://youtu.be/RiL-eLdGC9U")
+loader=youtube.YoutubeLoader.from_youtube_url("https://youtu.be/3Qkkxp48284")
 docs=loader.load()
 print(docs)
 with io.open("transcript.txt","w",encoding="utf-8")as f1:

tutorial60/main.py

@@ -0,0 +1,16 @@
+from langchain_community.document_loaders import PyPDFLoader
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+from langchain.vectorstores import Chroma 
+from langchain_openai import OpenAIEmbeddings
+import os
+
+loader=PyPDFLoader("vedas.pdf")
+docs=loader.load()
+text_splitter=RecursiveCharacterTextSplitter(chunk_size=1000,chunk_overlap=50)
+chunked_documents=text_splitter.split_documents(docs)
+vectordb=Chroma.from_documents(
+    documents=chunked_documents,
+    embedding=OpenAIEmbeddings(),
+    persist_directory=os.getcwd()+"/vector_index"
+)
+vectordb.persist()

tutorial60/prompt.txt

@@ -0,0 +1,19 @@
+"""
+You are very smart AI assistant who is very good in explaining and expressing answer for anything asked by the user
+user will ask question as {question}
+you must give answer with detailed explanation with bullets, heading , subheading etc. based on the {context} only
+always rename the title ,headings, subheaing in your answer but keep the meaning same.
+you must follow this  very stricty , dont use anything else other than the given {context}
+if no related information found from the {context} just reply with "I dont know", this is very important
+answer:
+"""
+
+
+"""
+You are very smart AI assistant who is very good in explaining and expressing answer for anything asked by the user
+user will ask question as {question}
+you must give answer with detailed explanation  based on the {context} only
+you must follow this  very stricty , dont use anything else other than the given {context}
+if no related information found from the {context} just reply with "I dont know", this is very important
+answer:
+"""

tutorial60/rag.py

@@ -0,0 +1,37 @@
+from langchain_community.vectorstores import Chroma 
+from langchain_openai import ChatOpenAI,OpenAIEmbeddings
+from langchain.chains import LLMChain
+from langchain.prompts import PromptTemplate
+import streamlit as st
+import os
+st.title("TALK TO ME")
+st.write("promting example with rag")
+
+vectordb=Chroma(persist_directory=os.getcwd()+"/vector_index",embedding_function=OpenAIEmbeddings())
+prompt_template="""
+You are very smart AI assistant who is very good in explaining and expressing answer for anything asked by the user
+user will ask question as {question}
+you must give answer with detailed explanation with bullets, heading , subheading etc. based on the {context} only
+always rewrite  the title ,headings, subheading in your ,try to avoid section numbers chapter numbers in the title or heading dont keep the same as the given in the {context} but keep the meaning same.
+you must follow this  very stricty , dont use anything else other than the given {context}
+if no related information found from the {context} just reply with "I dont know", this is very important
+answer:
+answer:
+"""
+llm=ChatOpenAI(model="gpt-4-turbo",max_tokens=1024)
+qa_chain=LLMChain(llm=llm,prompt=PromptTemplate.from_template(prompt_template))
+question=st.text_area("ask your question")
+
+if question is not None:
+    button=st.button("submit")
+    if button:
+        rd=vectordb.similarity_search(question,k=12)
+        context=""
+        for d in rd:
+            context +=d.page_content
+        response=qa_chain.invoke({"question":question,"context":context})
+        result=response["text"]
+        st.write(result)
+
+
+