Accessing Remote Jupyter at USC ISI

This guide provides instructions for accessing a Jupyter Notebook on a remote server and configuring a GPU for machine learning tasks using TensorFlow and PyTorch.

Accessing the Remote Jupyter Notebook

Prerequisites

• VPN connection, if required.
• SSH access to the server.
• Your server username (e.g., scho@bdnf.isi.edu).

Steps to Access

1. Open Terminal:

   • Windows: Use PowerShell or PuTTY.
   • macOS/Linux: Use the Terminal app.

2. SSH Connection:

   • Command:

     ssh -L 8080:localhost:8080 scho@bdnf.isi.edu
   • Replace scho with your server username.

3. Navigate to Work Directory:

   • Use cd to go to your directory:

     cd path/to/work/directory

4. Start Jupyter Notebook:

   • Run:

     jupyter lab --no-browser --port=8080
   • Copy the provided URL.

5. Access Notebook Locally:

   • Paste the URL into your local browser.

Setting Up GPU for TensorFlow

Prerequisites

• TensorFlow and other packages installed.
• Knowledge of available GPU numbers (0 to 7).

Steps for TensorFlow

1. Check GPU Availability:

   • In Jupyter, run:

     !nvidia-smi

2. Configure GPU in Notebook:

   • After loading packages, set GPU:

     import os
     import tensorflow as tf
     
     os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
     os.environ["CUDA_VISIBLE_DEVICES"] = "0"  # Choose an available GPU
     
     physical_devices = tf.config.list_physical_devices('GPU')
     print("Num GPUs Available: ", len(physical_devices))
     
     if physical_devices:
         tf.config.experimental.set_memory_growth(physical_devices[0], True)

Setting Up GPU for PyTorch

Prerequisites

• PyTorch installed.
• Knowledge of available GPU numbers.

Steps for PyTorch

1. Check GPU Availability:

   • Same as for TensorFlow, use !nvidia-smi.

2. Configure GPU in Notebook:

   • PyTorch automatically uses available GPUs, but you can specify one:

     import torch
     
     # Check if CUDA is available
     if torch.cuda.is_available():
         device = torch.device("cuda:0")  # Replace 0 with your GPU number
         print("Using GPU:", torch.cuda.get_device_name(0))
     else:
         device = torch.device("cpu")
         print("Using CPU")

   • Use device to move tensors or models to the selected device:

     model.to(device)

Notes

• For TensorFlow, CUDA_VISIBLE_DEVICES sets the specific GPU.
• For PyTorch, torch.device is used to specify the GPU.
• Always check GPU availability and usage before selecting one.