genai_bot.py

# %%capture
# !pip install -q transformers einops accelerate langchain bitsandbytes

!huggingface-cli login

!mkdir pdfs

# %%capture
# pip install pypdf

# %%capture
# pip install langchain-community

from langchain.document_loaders import PyPDFDirectoryLoader

loader=PyPDFDirectoryLoader("pdfs")

data=loader.load()

# we are going to create the chunks
from langchain.text_splitter import RecursiveCharacterTextSplitter

text_chunks=RecursiveCharacterTextSplitter(chunk_size=300,chunk_overlap=20)

chunks=text_chunks.split_documents(data)

len(chunks)

# creating embeddings using huggingface embedding model
from langchain.embeddings import HuggingFaceEmbeddings

# Commented out IPython magic to ensure Python compatibility.
# %%capture
# pip install sentence-transformers

embeddings=HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")

# %%capture
# pip install faiss-cpu

from langchain.vectorstores import FAISS

vectordabase=FAISS.from_documents(chunks,embeddings)

question="what is generative ai?"

vectordabase.similarity_search(question)

model="google/flan-t5-large"

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM,pipeline
from langchain import HuggingFacePipeline

tokenizer=AutoTokenizer.from_pretrained(model) # tokenization(token)
model1=AutoModelForSeq2SeqLM.from_pretrained(model) # (generation)

pipe=pipeline("text2text-generation",model=model1,tokenizer=tokenizer)

llm=HuggingFacePipeline(pipeline=pipe, model_kwargs={"temperature": 0, "max_length": 512})

# retrieval qa takes prompt template and llm
from langchain.prompts import PromptTemplate

template = """use the context to provide a concise answer and if you don't know just say don't now.
{context}
Question: {question}
Helpful Answer:"""
QA_CHAIN_PROMPT = PromptTemplate.from_template(template)

from langchain.chains import RetrievalQA
qa_chain = RetrievalQA.from_chain_type(
    llm, retriever=vectordabase.as_retriever(), chain_type_kwargs={"prompt": QA_CHAIN_PROMPT}
)

qa_chain.run("what is generative ai?")

question="more detail about generative ai?"
result = qa_chain({"query": question})
print(result["result"])

def fetch(question,history):
  result=qa_chain({"query":question})
  return result["result"]

# %%capture
# pip install gradio

import gradio as gr

gr.ChatInterface(
    fetch,
    chatbot=gr.Chatbot(height=300),
    textbox=gr.Textbox(placeholder="You can ask a question", container=False, scale=7),
    title="A Generative AI chatbot",
    description="You can ask questions related to Generative AI",
    theme="soft",
    examples=["what is Generative AI?", "what are the examples?", "what is LLM?"],
    cache_examples=True,
    retry_btn=None,
    undo_btn="Delete Previous",
    clear_btn="Clear",
).launch(share=True)