Fine-Tune with Google Colab

Free Jupyter notebooks with GPU access! No local setup required - perfect for learning and experimentation.

📋 Prerequisites

📦

1. Dataset Ready

Export your dataset from the EdukaAI application Export page. You'll upload it to Colab.

Open the EdukaAI app and go to Export.

👤

2. Google Account

Any Google account (Gmail) gives you free access to Colab.

💡 What is Google Colab?

Google Colab is a free cloud service that provides Jupyter notebooks with access to GPUs. You write and run Python code in your browser - no installation needed!

✅ Free

No cost to use, includes free T4 GPU

✅ No Setup

Everything runs in your browser

✅ GPU Included

Free Tesla T4 GPU (12GB VRAM)

1 Open Google Colab

1.Go to colab.research.google.com
2.Sign in with your Google account
3.Click "New Notebook"

2 Enable GPU (Important!)

⚠️ You MUST enable GPU for training!

1. Click Runtime in the menu
2. Select Change runtime type
3. Choose T4 GPU from the dropdown
4. Click Save

✅ Verify GPU: Run this in a code cell:

!nvidia-smi

You should see "Tesla T4" in the output

3 Upload Your Dataset

Upload your exported EdukaAI dataset to Colab:

# Upload your dataset
from google.colab import files
uploaded = files.upload()

# The file will be uploaded to the current directory
# You'll see the filename printed after upload

💡 Alternative: You can also mount Google Drive and load the file from there:

from google.colab import drive drive.mount('/content/drive')

4 Run Fine-Tuning

Here's a simple training template using Hugging Face libraries:

# Install libraries
!pip install transformers datasets peft accelerate

# Load libraries
from transformers import AutoModelForCausalLM, AutoTokenizer, TrainingArguments
from datasets import load_dataset
from peft import LoraConfig, get_peft_model

# Load base model
model = AutoModelForCausalLM.from_pretrained(
    "TinyLlama/TinyLlama-1.1B-Chat-v1.0",
    load_in_4bit=True  # Use 4-bit to save memory
)
tokenizer = AutoTokenizer.from_pretrained("TinyLlama/TinyLlama-1.1B-Chat-v1.0")

# Load your EdukaAI dataset
dataset = load_dataset('json', data_files='your_exported_file.json')

# Setup LoRA
lora_config = LoraConfig(
    r=8,
    lora_alpha=16,
    target_modules=["q_proj", "v_proj"],
    lora_dropout=0.05,
    bias="none",
    task_type="CAUSAL_LM"
)
model = get_peft_model(model, lora_config)

# Training arguments
training_args = TrainingArguments(
    output_dir="./results",
    num_train_epochs=1,
    per_device_train_batch_size=1,
    learning_rate=2e-4,
    logging_steps=10,
)

# ... (training code continues)

📚 Pre-made Templates

Search for these on Hugging Face or in Colab's examples:

• "LLM fine-tuning with PEFT"
• "Alpaca training notebook"
• "LoRA fine-tuning tutorial"

5 Save and Download Your Model

# Save your fine-tuned model
model.save_pretrained("./my-finetuned-model")
tokenizer.save_pretrained("./my-finetuned-model")

# Download to your computer
from google.colab import files
!zip -r my-finetuned-model.zip my-finetuned-model
files.download("my-finetuned-model.zip")

⚠️ Free Tier Limits

⏱️ Time Limits

Free tier provides ~12 hours of GPU time per day. Idle sessions timeout after 90 minutes.

💾 Memory

T4 GPU has 12GB VRAM. Use smaller models (1B-3B parameters) or 4-bit quantization.

🔌 Session End

When you close the browser, the runtime resets. Always save results to Google Drive!

💡 Colab Tips

💾 Save to Drive

Mount Google Drive to save models between sessions.

🔋 Prevent Timeout

Keep the browser tab active, or use a JavaScript snippet to prevent idle timeout.