In this session, you will learn how to upload your fine-tuned HuggingFace model and deploy it using two different platforms: Streamlit and Gradio. First, you will learn how to properly upload your model to the HuggingFace Hub, a platform that allows you to easily share and access models. Then, you will learn how to use Streamlit and Gradio, two user-friendly platforms, to deploy your model and make it accessible to others. By the end of this session, you will be able to confidently upload and deploy your models for use by others.
First, it logs in to the Hugging Face Hub using notebook_login(), which allows for easy uploading and access to Hugging Face models.
#Importing the necessary libraries
from huggingface_hub import notebook_login
from transformers import pipeline
#Login to Hugging Face account from the Jupyter notebook
notebook_login()
Next, it pushes the trainer object, model object, and dataset object to the Hub using their push_to_hub() methods. This uploads these objects to the Hub for others to use and access.
#Pushing the trained model, trainer and dataset to the Hugging Face Hub
trainer.push_to_hub('Your Fine-Tuned HuggingFace Model')
model.push_to_hub('Your Fine-Tuned HuggingFace Model')
dataset.push_to_hub('Your Fine-Tuned HuggingFace Model')
Finally, the code loads a pre-trained Danish emotion classification model using model_ckpt = ‘Your Fine-Tuned HuggingFace Model’, and creates a pipeline using the pipeline() method from the transformers library. This pipeline can then be used to classify text inputs into one of several emotional categories.
#Defining the path to the trained model checkpoint on the Hugging Face Hub
model_ckpt = 'Your Fine-Tuned HuggingFace Model'
#Creating a pipeline for example text classification using the fine-tuned model from the Hugging Face Hub
pipe = pipeline('text-classification', model=model_ckpt)
#Passing the Danish text examples to the pipeline and getting the predicted labels
pipe(['Du er en idiot!', 'Jeg er sulten!'])
You can also deploy your fine-tuned Hugging Face model on Streamlit through the Hugging Face web interface. Here are the steps to do so:
That’s it! Your fine-tuned Hugging Face model is now deployed on Streamlit through the Hugging Face web interface and can be accessed using your app.
1. Danish text sentiment analysis using Gradio:
# app.py
import gradio as gr
from transformers import pipeline
from transformers import AutoTokenizer, AutoModelForSequenceClassification
#Defining the classify function which takes text as input and returns the label of the sentiment
def classify(text):
# Initializing the pipeline for sentiment analysis
cls = pipeline('text-classification', model='RJuro/dk_emotion_bert_in_class')
# Predicting the sentiment label for the input text
return cls(text)[0]['label']
#Creating the Gradio interface with input textbox and output text
gr.Interface(fn=classify, inputs=["textbox"], outputs="text").launch()
# requirements.txt
transformers
gradio
torch
2. Hugging Face Sentiment Analysis Spaces Example using Streamlit:
# app.py
import streamlit as st
from transformers import pipeline
from textblob import TextBlob
pipe = pipeline('sentiment-analysis')
st.title("Hugging Face Sentiment Analysis Spaces Example")
st.subheader("What framework would you like to use for Sentiment Analysis")
#Picking what NLP task you want to do
option = st.selectbox('Framework',('Transformers', 'TextBlob')) #option is stored in this variable
#Textbox for text user is entering
st.subheader("Enter the text you'd like to analyze.")
text = st.text_input('Enter text') #text is stored in this variable
if option == 'Transformers':
out = pipe(text)
else:
out = TextBlob(text)
out = out.sentiment
st.write("Sentiment of Text: ")
st.write(out)
# requirements.txt
torch
transformers
textblob
Build an exctiting and perhapps also fun application using techniques learned in this module Submission: Github repo as usual.