app.py

import gradio as gr
from langchain.document_loaders import PDFMinerLoader, PyMuPDFLoader
from langchain.text_splitter import CharacterTextSplitter
import chromadb
import chromadb.config
from chromadb.config import Settings


import gradio as gr

def get_context(query_text):
    query_emb = st_model.encode(query_text)
    query_response = collection.query(query_embeddings=query_emb.tolist(), n_results=4)
    context = query_response['documents'][0][0]
    context = context.replace('\n', ' ').replace('  ', ' ')
    return context

def local_query(query, context):
    t5query = """Using the available context, please answer the question. 
    If you aren't sure please say i don't know.
    Context: {}
    Question: {}
    """.format(context, query)
    
    inputs = tokenizer(t5query, return_tensors="pt")
    outputs = model.generate(**inputs, max_new_tokens=20)
    
    return tokenizer.batch_decode(outputs, skip_special_tokens=True)

def run_query(query):
    context = get_context(query)
    result = local_query(query, context)
    return result


    
def upload_pdf(file):
    # Save the uploaded file
    file_name = file.name
    pdf_filename = os.path.basename(file_path)

    # Load a document
    loader = PDFMinerLoader(pdf_filename)
    doc = loader.load()

    text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
    texts = text_splitter.split_documents(doc)

    texts = [i.page_content for i in texts]

    doc_emb = st_model.encode(texts)
    doc_emb = doc_emb.tolist()

    ids = [str(uuid.uuid1()) for _ in doc_emb]

    client = chromadb.Client()
    # Create collection. get_collection, get_or_create_collection, delete_collection also available!
    collection = client.create_collection("test_db") 
    
    collection.add(
        embeddings=doc_emb,
        documents=texts,
        ids=ids
    )

    return run_query("how to reduce waste?")


from transformers import T5ForConditionalGeneration, AutoTokenizer
import torch


model_name = 'google/flan-t5-base'

model = T5ForConditionalGeneration.from_pretrained(model_name, device_map='auto', offload_folder="offload")
tokenizer = AutoTokenizer.from_pretrained(model_name)

import uuid
from sentence_transformers import SentenceTransformer

ST_name = 'sentence-transformers/sentence-t5-base'

st_model = SentenceTransformer(ST_name)



iface = gr.Interface(
    fn=upload_pdf,
    inputs="file",
    outputs="text",
    title="PDF File Uploader",
    description="Upload a PDF file and get its filename.",
)

iface.launch()