# Quickstart

# Getting an Access to MoAI Platform

Please obtain a container or virtual machine on the MoAI Platform from your infrastructure provider and follow the instructions to connect via SSH. For example, you can apply for a trial container on the MoAI Platform or use public cloud services based on the MoAI Platform.

MoAI Platform Trial Container (Inquiries: support@moreh.io)
KT Cloud's Hyperscale AI Computing (https://cloud.kt.com/solution/hyperscaleAiComputing/)

# Verfying MoAI Accelerator

To train models like the sLLM introduced in this tutorial, it's important to select an appropriate size of the MoAI Accelerator. First, use the moreh-smi command to check the currently used MoAI Accelerator.

$ moreh-smi
+---------------------------------------------------------------------------------------------------+
|                                                 Current Version: 24.11.0  Latest Version: 24.11.0 |
+---------------------------------------------------------------------------------------------------+
|  Device  |        Name         |      Model    |  Memory Usage  |  Total Memory  |  Utilization   |
+===================================================================================================+
|  * 0     |   MoAI Accelerator  |  Large.256GB  |  -             |  -             |  -             |
+---------------------------------------------------------------------------------------------------+

# Verifying PyTorch Installation

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
...

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 Installation part below.

# Installation (if required)

# Setting up an Environment with Anaconda

To begin training, first create a conda environment:
```
conda create --name <my-env> python=3.8
```
Replace <my-env> with your desired environment name.
Activate the conda environment:
```
conda activate <my-env>
```
The MoAI Platform supports various PyTorch versions, allowing you to choose the one that fits your needs.
```
pip install torch==1.13.1+moreh24.11.209
```

# Running a Sample Code

Clone the repository which includes example codes

git clone https://github.com/moreh-dev/quickstart
cd quickstart

Install the dependency packages

pip install -r requirements/requirements_llama3.txt

Run a training script

python tutorial/train_llama3.py\
    --epochs 1\
    --batch-size 256\
    --block-size 1024\
    --lr 0.00001

You can see logs like the one below.

| INFO     | __main__:main:242 - Model load and warmup done. Duration: 316.09
| INFO     | __main__:main:252 - [Step 5/1121] | Loss: 1.9141 | Duration: 32.67 sec | Throughput: 32100.30 tokens/sec
| INFO     | __main__:main:252 - [Step 10/1121] | Loss: 1.9297 | Duration: 40.53 sec | Throughput: 32337.05 tokens/sec
| INFO     | __main__:main:252 - [Step 15/1121] | Loss: 1.9297 | Duration: 40.80 sec | Throughput: 32129.29 tokens/sec
| INFO     | __main__:main:252 - [Step 20/1121] | Loss: 1.9062 | Duration: 40.06 sec | Throughput: 32715.97 tokens/sec
| INFO     | __main__:main:252 - [Step 25/1121] | Loss: 1.9062 | Duration: 40.14 sec | Throughput: 32656.86 tokens/sec
| INFO     | __main__:main:252 - [Step 30/1121] | Loss: 1.8672 | Duration: 40.36 sec | Throughput: 32477.78 tokens/sec

tutorial