Week 3 Practical Guide - Model Deployment

⚠️ Known Issues & Options

There are version compatibility issues between Hugging Face and Kaggle/Colab. If you encounter errors:

Option 1 (Recommended): Follow tasks below using Hugging Face Spaces

Option 2: Run everything on Kaggle (see example: dmitrykonovalov/fastai22-lesson2-prac3-app-250605)

Option 3: Install Gradio locally on your laptop (ask ChatGPT: "how to install and run gradio locally on windows-11")

Note: Hugging Face may be blocked on JCU network. Use personal hotspot if needed.

TASK 1: Your First Gradio App (Hello World)

1 Create Hugging Face Account

Go to huggingface.co/spaces
Sign up for free account
Log in

2 Create New Space

Click "Create new Space"
Name: fastai-lesson2-hello (or your choice)
Select: Gradio as SDK
License: Apache 2.0
Visibility: Public
Click "Create Space"

3 Create app.py File

Click "Files" tab
Click "+ Add file" → "Create a new file"
Filename: app.py
Copy this code:

import gradio as gr

def greet(name):
    return "Hello " + name + "!!"

demo = gr.Interface(fn=greet, inputs="text", outputs="text")
demo.launch()

Click "Commit new file to main"

4 Test Your App

Wait 1-2 minutes for build to complete
Your app will appear in the Space
Type your name in the text box
Click Submit - you should see "Hello [your name]!!"

✓ SUBMIT: URL of your Hugging Face Space (or screenshot if running locally)

TASK 2: Deploy a Cat vs Dog Classifier

Part A: Train and Export Model

1 Open Training Notebook

Go to Kaggle
Copy this notebook: dmitrykonovalov/fastai-lesson2-part2b-train-save-model-v250528
Or use your Week 2 cat/dog classifier

2 Add Export Code

At the end of your training code, add:

learn.export('model.pkl')  # Save the trained model

Run all cells
Save the notebook

3 Download model.pkl

Click "Output" tab in Kaggle
Find model.pkl
Download it to your computer

Part B: Deploy the Model

4 Duplicate Example Space

Go to: huggingface.co/spaces/dmitryakonovalov/fastai-lesson2-part2-cats-dogs-v2
Click "⋮" (three dots) in top-right
Select "Duplicate this Space"
Give it a new name: your-username/cat-dog-classifier

6 Fix app.py

Click on app.py to edit it
Find this line:

learn = load_learner('my_model.pkl')  # This will give error!

Change it to:

learn = load_learner('model.pkl')  # Use your uploaded model

Commit changes

TASK 3: Deploy Your Three-Bear Classifier

Remember Week 2? You trained a model to recognize grizzly, black, and teddy bears. Now deploy it!

1 Export Your Bear Model

Open your Week 2 bear classifier notebook
Add at the end: learn.export('bear_model.pkl')
Run and download bear_model.pkl

2 Duplicate Your Task 2 Space

Go to your cat-dog classifier Space from Task 2
Click "⋮" → "Duplicate this Space"
Name it: your-username/bear-classifier

3 Replace the Model

Delete old model.pkl
Upload your bear_model.pkl
Rename it to model.pkl

4 Update app.py

Edit app.py
Change title from "Cat Dog Classifier" to "Bear Classifier"
Change description to mention the three bear types
Example:

demo = gr.Interface(
    fn=classify_image,
    inputs=gr.Image(),
    outputs=gr.Label(num_top_classes=3),
    title="🐻 Bear Classifier",
    description="Upload a bear image to classify: Grizzly, Black, or Teddy Bear"
)

5 Test with Bear Images

Search Google Images for bears
Test: grizzly bear, black bear, teddy bear
Check if predictions match

✓ SUBMIT: URL of your Bear Classifier Space

Common Errors & Solutions

Error	Solution
`FileNotFoundError: model.pkl`	Make sure model.pkl is uploaded AND filename in app.py matches exactly
`ModuleNotFoundError: No module named 'fastai'`	Create requirements.txt with: `fastai gradio`
Build keeps failing	Check Logs tab for specific error. Try Option 2 (Kaggle) or Option 3 (local)
Space shows blank/loading forever	Check if HF is blocked on network. Try personal hotspot or home internet
Version mismatch errors	Pin versions in requirements.txt: `fastai==2.7.14 gradio==4.0.0`

✓ Final Checklist

Before You Leave:

☐ Task 1: Hello World app working
☐ Task 2: Cat/Dog classifier deployed
☐ Task 3: Bear classifier deployed
☐ All Spaces are set to "Public"
☐ Tested each app with multiple images
☐ Submitted all URLs to LearnJCU

Bonus Challenges (Optional):