pdf_qa / app.py
Adarsh-aot's picture
Update app.py
228558e verified
raw
history blame
3.26 kB
import streamlit as st
import csv
import chromadb
from chromadb.utils import embedding_functions
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
from transformers import pipeline
from langchain.llms import HuggingFacePipeline
# Load sample data (a restaurant menu of items)
# with open('./data.csv') as file:
# lines = csv.reader(file)
# # Store the name of the menu items in this array. In Chroma, a "document" is a string i.e. name, sentence, paragraph, etc.
# documents = []
# # Store the corresponding menu item IDs in this array.
# metadatas = []
# # Each "document" needs a unique ID. This is like the primary key of a relational database. We'll start at 1 and increment from there.
# ids = []
# id = 1
# # Loop thru each line and populate the 3 arrays.
# for i, line in enumerate(lines):
# if i == 0:
# # Skip the first row (the column headers)
# continue
# documents.append(line[0])
# metadatas.append({"item_id": line[1]})
# ids.append(str(id))
# id += 1
# Instantiate chromadb instance. Data is stored on disk (a folder named 'my_vectordb' will be created in the same folder as this file).
chroma_client = chromadb.PersistentClient(path="vector_db")
# Select the embedding model to use.
# List of model names can be found here https://www.sbert.net/docs/pretrained_models.html
sentence_transformer_ef = embedding_functions.SentenceTransformerEmbeddingFunction(model_name="all-mpnet-base-v2")
# Use this to delete the database
# chroma_client.delete_collection(name="my_collection")
# Create the collection, aka vector database. Or, if database already exist, then use it. Specify the model that we want to use to do the embedding.
collection = chroma_client.get_or_create_collection(name="my_collection", embedding_function=sentence_transformer_ef)
# Add the data to the collection
# collection.add(
# documents=documents,
# metadatas=metadatas,
# ids=ids
# )
# Streamlit app layout
st.title("ChromaDB and HuggingFace Pipeline Integration")
query = st.text_input("Enter your query:", value="director")
if st.button("Search"):
results = collection.query(
query_texts=[query],
n_results=1,
include=['documents', 'distances', 'metadatas']
)
st.write("Query Results:")
st.write(results['metadatas'])
if results['documents']:
context = results['documents']
st.write("Context:")
st.write(context)
tokenizer = AutoTokenizer.from_pretrained("MBZUAI/LaMini-T5-738M")
model = AutoModelForSeq2SeqLM.from_pretrained("MBZUAI/LaMini-T5-738M")
pipe = pipeline(
"text2text-generation",
model=model,
tokenizer=tokenizer,
max_length=512
)
local_llm = HuggingFacePipeline(pipeline=pipe)
l = f"""
Use the following pieces of context to answer the question at the end. If you don't know the answer, just say that you don't know, don't try to make up an answer.
{context}
Question: {query}
Helpful Answer:
"""
answer = local_llm(l)
st.write("Answer:")
st.write(answer)