Skip to main content

Run Hugging Face Models in Google Colab

Overview​

Google Colab and Hugging Face have added an “Open in Colab” button to all model cards on the Hub. With a single click, you can generate a pre-configured notebook to load, test, and even fine-tune any model—skipping boilerplate and accelerating experiments. (medium.com)

Prerequisites​

  • Google account to access Colab
  • Hugging Face account (optional, but recommended for saving tokens and private models)
  • An up-to-date browser
GOOD TO KNOW

If you are a model author, simply include a notebook.ipynb file in the root of the repository, and it will be used instead of the automatically generated notebook. (medium.com)

Step-by-Step​

1. Choose the model​

Go to any model card on the Hub, for example:

https://huggingface.co/google/gemma-3-27b-it

2. Open in Colab​

Click on Use this model → Open in Colab or simply add /colab to the end of the URL:

https://huggingface.co/google/gemma-3-27b-it/colab

3. Configure the environment​

In the generated notebook, go to Runtime ▸ Change runtime type and select GPU (or TPU if available).

Free GPUs in Colab are sufficient for inference on most Transformer-based models.

4. Execute the cells​

The first cell usually installs dependencies:

!pip install -q transformers accelerate

Next, try out the model:

from transformers import pipeline
pipe = pipeline("text-generation", model="google/gemma-3-27b-it")
print(pipe("Hello, how are you?")[0]['generated_text'])

5. Customize​

  • Replace model= with another Hugging Face model ID
  • Adjust parameters like max_new_tokens, temperature, top_p
  • Perform fine-tuning by adding your own training step
pip install huggingface_hub
export MODEL_ID=google/gemma-3-27b-it
python - <<'PY'
from huggingface_hub import InferenceClient
client = InferenceClient(model="$MODEL_ID")
print(client.text_generation("Hello, world!"))
PY

For Model Authors​

  1. Create or upload a notebook.ipynb file demonstrating advanced usage of your model.
  2. Commit it to the same repository; the Hub will prioritize this file over the auto-generated one.
  3. The “Open in Colab” button will now point to your custom notebook.

Next Steps​

Resources​

Tutorial created on June 8, 2025, based on the Google Colab article.