# 1. Preparing for Fine-tuning

Setting up the PyTorch execution environment on the MoAI Platform is similar to setting it up on a typical GPU server.
For a smooth tutorial experience, the following specifications are recommended:

• CPU: 16 cores or more

• Memory: 256GB or more

• MAF version: 24.11.0

• Storage: 40GB or more

Please verify that your environment meets these requirements before starting the tutorial.

After connecting to the container via SSH, run the following command to check if PyTorch is installed in the current conda environment:

$ conda list torch
...
# Name                    Version                   Build  Channel
torch                     2.1.0+cu118.moreh24.11.0          pypi_0    pypi
...

The version name includes both the PyTorch version and the MoAI version required to run it.
In the example above, it indicates that PyTorch 2.1.0+cu118 is installed with MoAI version 24.11.0.

If you see the message conda: command not found, if the torch package is not listed, or if the torch package exists but does not include "moreh" in the version name, please follow the instructions in the Prepare Fine-tuning on MoAI Platform document to create a conda environment.

If the moreh version is not 24.11.0 but a different version, please execute the following code.

$ update-moreh --target 24.11.0 --torch 2.1.0
Currently installed: 24.9.0
Possible upgrading version: 24.11.0
  
Do you want to upgrade? (y/n, default:n)
y

Run the following command to confirm that the torch package is properly imported and the MoAI Accelerator is recognized.

$ python
Python 3.8.20 (default)
[GCC 11.2.0] :: Anaconda, Inc. on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
...
>>> torch.cuda.device_count()
1
>>> torch.cuda.get_device_name()
[info] Requesting resources for MoAI Accelerator from the server...
[info] Initializing the worker daemon for MoAI Accelerator
[info] [1/1] Connecting to resources on the server (192.168.xxx.xx:xxxxx)...
[info] Establishing links to the resources...
[info] MoAI Accelerator is ready to use.
'MoAI Accelerator'
>>> quit()

Execute the following command to download the PyTorch script for training from the GitHub repository.
In this tutorial, we will be using the train_llama3.py script located inside the tutorial directory.

$ sudo apt-get install git
$ git clone https://github.com/moreh-dev/quickstart.git
$ cd quickstart
~/quickstart$ ls tutorial
...  train_llama3.py  ...

Execute the following command to install third-party Python packages required for script execution:

$ pip install -r requirements/requirements_llama3.txt

To access and download the Llama3 8B model checkpoint from Hugging Face Hub, you will need to agree to the community license and provide your Hugging Face token information.

First, enter the required information and agree to the license on the following site.

After submitting the agreement, confirm that the page status has changed as shown below.

Now you can authenticate your Hugging Face token with the following command:

huggingface-cli login