Colab Features & Tips

Colab Features & Tips

Colab Forms

Create interactive UI elements with #@param annotations:

#@title Configuration
learning_rate = 0.001 #@param {type:"number"}
epochs = 10 #@param {type:"slider", min:1, max:100}
model_type = "ResNet50" #@param ["ResNet50", "VGG16", "EfficientNet"]
use_augmentation = True #@param {type:"boolean"}
description = "My experiment" #@param {type:"string"}

print(f"Training {model_type} for {epochs} epochs at lr={learning_rate}")

Secrets (API Keys)

Store sensitive values securely — not in notebook code:

# Access via Colab Secrets (left sidebar → key icon)
from google.colab import userdata

api_key = userdata.get('OPENAI_API_KEY')
hf_token = userdata.get('HF_TOKEN')

Never hardcode API keys in notebooks — they get committed to GitHub.

Colab AI (Gemini Integration)

Colab has built-in AI assistance powered by Gemini:

• Ctrl+I — open AI chat panel
• Alt+Enter — generate code from comment
• Explain code, fix errors, generate cells

# Type a comment and press Alt+Enter to generate code
# Load a CSV file and show basic statistics

GitHub Integration

# Clone a repository
!git clone https://github.com/username/repo.git
%cd repo

# Or open directly from GitHub
# File → Open notebook → GitHub tab → paste URL

# Commit changes back
!git config --global user.email "you@example.com"
!git config --global user.name "Your Name"
!git add .
!git commit -m "Update notebook"
!git push

Installing Packages

# Install for current session
!pip install transformers accelerate datasets

# Install specific version
!pip install torch==2.1.0

# Install from GitHub
!pip install git+https://github.com/huggingface/transformers.git

# Conda (slower but available)
!pip install condacolab
import condacolab
condacolab.install()
!conda install -c conda-forge rdkit

Note: Installations are lost when the session ends. Add installs at the top of your notebook.

Persistent Storage Patterns

# Option 1: Google Drive (recommended)
from google.colab import drive
drive.mount('/content/drive')
SAVE_PATH = '/content/drive/MyDrive/my_project/'

# Option 2: Download results before session ends
from google.colab import files
files.download('model_weights.pth')

# Option 3: Upload to cloud storage
from google.cloud import storage
# ... GCS upload code

# Option 4: Push to GitHub
!git push origin main

Colab with Hugging Face

!pip install transformers datasets accelerate

from transformers import pipeline

# Text generation
generator = pipeline('text-generation', model='gpt2')
result = generator("The future of AI is", max_length=50)
print(result[0]['generated_text'])

# Image classification
classifier = pipeline('image-classification', model='google/vit-base-patch16-224')

# Use HF token for gated models
from huggingface_hub import login
from google.colab import userdata
login(token=userdata.get('HF_TOKEN'))

Colab with OpenAI

!pip install openai

from openai import OpenAI
from google.colab import userdata

client = OpenAI(api_key=userdata.get('OPENAI_API_KEY'))

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Explain transformers in 3 sentences"}]
)
print(response.choices[0].message.content)

Sharing & Collaboration

• Share button — share like a Google Doc (view/comment/edit)
• File → Save a copy in GitHub — save to a repo
• File → Download → .ipynb — download notebook
• Published notebooks: File → Publish to GitHub Gist

Colab Pro Tips

# Prevent session timeout (run in browser console)
# Open browser DevTools → Console → paste:
function ClickConnect(){
  console.log("Keeping alive...");
  document.querySelector("colab-toolbar-button#connect").click()
}
setInterval(ClickConnect, 60000)

# Check remaining GPU quota
# Runtime → View resources

# Use background execution (Pro+)
# Runtime → Run all → close browser tab

# Faster I/O — copy data to /content (local SSD) first
!cp /content/drive/MyDrive/large_dataset.zip /content/
!unzip /content/large_dataset.zip -d /content/data/

Environment Variables

import os

# Set for current session
os.environ['MY_VAR'] = 'value'

# Or use Colab Secrets (preferred for sensitive values)
from google.colab import userdata
os.environ['API_KEY'] = userdata.get('API_KEY')

Colab Limitations

• Session resets after inactivity (~90 min) or max runtime (12h free)
• No persistent filesystem (use Drive or download)
• GPU availability not guaranteed on free tier
• Limited to one GPU per session
• No custom domain or public URL (use ngrok for that)

# Expose a local server publicly with ngrok
!pip install pyngrok
from pyngrok import ngrok

# Start a tunnel
public_url = ngrok.connect(8501)
print(f"App URL: {public_url}")

# Run Streamlit app
!streamlit run app.py &