AI Project: Fine-Tuning a Hugging Face Model (Part 3: Evaluation & Sharing)

This is the final part of our fine-tuning series. In this article, we’ll explore Hugging Face Evaluate and Share to wrap up our journey.

• In Part 1, we prepared the data.
• In Part 2, we used the Trainer to build our custom model.

Now, let’s see how good our model is and share it with the world!

Step 1: Evaluate the Model

The Trainer object we built in Part 2 already has our “test” dataset. We just need to call the .evaluate() method.

# Assuming 'trainer' is your trained Trainer object from Part 2

print("Evaluating model...")
eval_results = trainer.evaluate()

print(eval_results)

This will run your model against the 25,000-row test dataset and give you a result like:

{'eval_loss': 0.123, 'eval_accuracy': 0.95, ...}

An accuracy of 0.95 means our model is 95% accurate at classifying movie reviews! We’ve successfully fine-tuned a general-purpose AI to be an expert.

Step 2: Log In to the Hub

To share your model, you need to log in to your Hugging Face account from your terminal.

# 1. Install the CLI tool
pip install huggingface_hub

# 2. Run the login command
huggingface-cli login

This will ask for an “Access Token.” You can generate one from your Hugging Face account settings (under “Access Tokens”).

Step 3: Push to Hub

Once you’re logged in, you can push your model with one command.

# Give your model a name on the hub
model_repo_name = "my-awesome-sentiment-model-v1"

print("Pushing model to the Hub...")
trainer.push_to_hub(model_repo_name)

That’s it! Your custom model is now saved on your Hugging Face profile. You (or anyone else) can now load it using the pipeline, just like any official model:

pipeline("sentiment-analysis", model="YourUsername/my-awesome-sentiment-model-v1")