Added files

README.md

@@ -1,12 +1,2 @@
----
-title: Retrieval Augmented Generation RAG
-emoji: 📈
-colorFrom: green
-colorTo: yellow
-sdk: streamlit
-sdk_version: 1.29.0
-app_file: app.py
-pinned: false
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# Retrieval-Augmented-Generation-RAG-
+Simple RAG using your own pdfs without any GPU!

app.py

@@ -0,0 +1,21 @@
+from retrieve import qa_chain, process_llm_response
+import streamlit as st
+
+
+def main():
+    qa = qa_chain()
+    st.title('NCERT-GPT')
+    text_query = st.text_area('Ask any question from NCERT 11th and 12th Chemistry Texts!')
+    generate_response_btn = st.button('Run RAG')
+    
+    st.subheader('Response')
+    if generate_response_btn and text_query is not None:
+        with st.spinner('Generating Response. Please wait...'):
+            text_response = qa(text_query)
+            if text_response:
+                st.write(text_response)
+            else:
+                st.error('Failed to get response')
+
+if __name__ == "__main__":
+    main()    

ingest.py

@@ -0,0 +1,43 @@
+#importing dependencies
+from langchain.embeddings import HuggingFaceBgeEmbeddings
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+from langchain.vectorstores import Chroma
+from langchain.document_loaders import PyPDFDirectoryLoader
+import time
+
+#loading data
+
+loader = PyPDFDirectoryLoader('data/')
+documents = loader.load()
+print(len(documents))
+
+#splitting
+
+splitter = RecursiveCharacterTextSplitter(chunk_size = 10000, chunk_overlap = 500)
+text_chunks = splitter.split_documents(documents)
+print(len(text_chunks))
+
+#loading HuggingFaceBGE embeddings
+
+model_name = "BAAI/bge-small-en"
+model_kwargs = {"device": "cpu"}
+encode_kwargs = {"normalize_embeddings": True}
+embeddings = HuggingFaceBgeEmbeddings(
+    model_name=model_name, model_kwargs=model_kwargs, encode_kwargs=encode_kwargs
+)
+
+print('Embeddings loaded!')
+
+# creating NCERT Textbooks vector database.
+
+t1 = time.time()
+persist_directory = 'dbname'
+vectordb = Chroma.from_documents(
+    documents = text_chunks,
+    embedding = embeddings,
+    collection_metadata = {"hnsw:space": "cosine"},
+    persist_directory = persist_directory
+)
+t2 = time.time()
+print('Time taken for building db : ', (t2 - t1))
+

requirements.txt

@@ -0,0 +1,10 @@
+accelerate
+chromadb
+huggingface-hub
+langchain
+pypdf
+sentence-transformers
+sentencepiece
+streamlit
+torch
+transformers

retrieve.py

@@ -0,0 +1,73 @@
+from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer, pipeline
+from langchain.llms import HuggingFaceHub, HuggingFacePipeline 
+from dotenv import load_dotenv
+from langchain.embeddings import HuggingFaceBgeEmbeddings
+from langchain.vectorstores import Chroma
+from langchain.chains import RetrievalQA
+import textwrap
+import os
+
+
+def load_vector_store():
+    model_name = "BAAI/bge-small-en"
+    model_kwargs = {"device": "cpu"}
+    encode_kwargs = {"normalize_embeddings": True}
+    embeddings = HuggingFaceBgeEmbeddings(
+        model_name=model_name, model_kwargs=model_kwargs, encode_kwargs=encode_kwargs
+    )
+    print('Embeddings loaded!')
+    load_vector_store = Chroma(persist_directory = 'vector stores/ncertdb', embedding_function = embeddings)
+    print('Vector store loaded!')
+
+    retriever = load_vector_store.as_retriever(
+        search_kwargs = {"k" : 2},
+    )
+    return retriever
+
+    
+#model 
+def load_model():  
+    load_dotenv()  
+    repo_id = 'llmware/bling-sheared-llama-1.3b-0.1'
+    llm = HuggingFaceHub(
+        repo_id = repo_id,
+        model_kwargs = {'max_new_tokens' : 100}
+    )
+    print(llm('HI!'))
+    return llm
+
+
+def qa_chain():
+    retriever = load_vector_store()
+    llm = load_model()
+    qa = RetrievalQA.from_chain_type(
+        llm = llm,
+        chain_type = 'stuff',
+        retriever = retriever,
+        return_source_documents = True,
+        verbose = True 
+    )
+    return qa
+
+def wrap_text_preserve_newlines(text, width=110):
+    # Split the input text into lines based on newline characters
+    lines = text.split('\n')
+
+    # Wrap each line individually
+    wrapped_lines = [textwrap.fill(line, width=width) for line in lines]
+
+    # Join the wrapped lines back together using newline characters
+    wrapped_text = '\n'.join(wrapped_lines)
+
+    return wrapped_text
+
+def process_llm_response(llm_response):
+    print(wrap_text_preserve_newlines(llm_response['result']))
+    print('\n\nSources:')
+    for source in llm_response["source_documents"]:
+        print(source.metadata['source'])
+
+qa = qa_chain()
+
+response = qa('What are types of Embedded system?')
+process_llm_response(response)