Jupyter Notebooks

What is a Jupyter Notebook?

A Jupyter Notebook is an open-source web application for creating and sharing documents containing:

• Live code

• Equations

• Visualizations

• Narrative text

It is a standard tool for data science, machine learning, and AI research.

---

Getting Started with Your Notebook

When you open your rented Jupyter environment, you'll see the Jupyter file browser. From here, you can:

• Create a new notebook:

  • Click New → Python 3 Kernel to start a fresh notebook.

• Upload files:

  • Use the Upload button to bring in datasets or existing notebooks.

---

Using GPU in Your Notebook

NVIDIA CUDA Environment

Your CUDA environment comes with PyTorch pre-installed.

• Verify GPU access:

  import torch
  print(f"CUDA available: {torch.cuda.is_available()}")
  print(f"GPU: {torch.cuda.get_device_name(0)}")
  print(f"VRAM: {torch.cuda.get_device_properties(0).total_mem / 1e9:.1f} GB")

• Load and run a model on GPU:

  from transformers import AutoModelForCausalLM, AutoTokenizer
  
  model_name = "meta-llama/Llama-3.2-1B"
  tokenizer = AutoTokenizer.from_pretrained(model_name)
  
  model = AutoModelForCausalLM.from_pretrained(
      model_name,
      torch_dtype=torch.float16,
      device_map="auto"
  )
  
  inputs = tokenizer("Hello, world!", return_tensors="pt").to("cuda")
  outputs = model.generate(**inputs, max_new_tokens=50)
  
  print(tokenizer.decode(outputs[0]))

---

Apple MLX Environment

Your MLX environment comes with Apple's ML framework pre-installed.

• Check the MLX backend:

  import mlx.core as mx
  print(f"MLX backend: {mx.default_device()}")

• Create arrays on the GPU:

  a = mx.random.normal((1000, 1000))
  b = mx.random.normal((1000, 1000))
  c = a @ b  # Matrix multiplication on Metal GPU
  mx.eval(c)
  print(f"Result shape: {c.shape}")

• Run language models with mlx-lm:

  from mlx_lm import load, generate
  
  model, tokenizer = load("mlx-community/Llama-3.2-1B-Instruct-4bit")
  response = generate(
      model,
      tokenizer,
      prompt="Explain quantum computing in simple terms:",
      max_tokens=200
  )
  
  print(response)

---

Installing Additional Packages

• Use pip directly in a notebook cell, e.g.

  !pip install datasets scikit-learn matplotlib seaborn

Remember: Installed packages are ephemeral and do not persist after your session ends. Include your pip install commands at the top of your notebook for future sessions.

Learn more about Jupyter Notebooks at the official docs: https://docs.jupyter.org/en/latest

Tips for Productive Sessions

• Save your work frequently. Download notebooks and outputs before your session expires.

• Monitor VRAM usage. Use !nvidia-smi (CUDA) or check memory in your code to avoid out-of-memory errors.

• Use efficient data types. Load models in float16 or int4 to maximize VRAM usage.

• Plan for ephemeral storage. Upload datasets at the start of each session and download results before ending.

• Extend if needed. If your training job requires more time, extend your rental before it expires to avoid interruptions.

Common Workflows

Workflow	Recommended GPU	Duration
Prototyping & Testing	Any available GPU	30 min – 1 hour
Fine-tuning small models	RTX 4070+ (12+ GB VRAM)	2 – 4 hours
Fine-tuning large models	RTX 4090 / Multi-GPU (24+ GB VRAM)	4 – 8 hours
Inference & Evaluation	Any GPU matching model requirements	30 min – 1 hour
Data Processing	Any GPU with sufficient VRAM	1 – 2 hours
MLX Fine-tuning or Inference	Apple M-series (16+ GB unified memory)	1 – 2 hours