Home / Getting started
Getting started

From quote to your first model in about an hour.

Everything you need to go from “we need GPUs” to a model answering requests. No special tools: just a terminal and the login details we email you.

1

Choose your setup

Not sure which GPU you need? Start with the workload guide, then build an estimate in the price calculator. As a rule of thumb: L40S or RTX PRO 6000 to serve models, H100 or H200 to fine-tune, B200 or B300 to train large models.

2

Get a quote and confirm

Click Request this quote in the calculator, or reserve capacity for a start date. We confirm availability and a firm price within . Nothing is charged until you accept the quote.

3

Create an SSH key

An SSH key lets you log in securely without a password. Run this once on your own laptop (Mac, Linux, or Windows PowerShell), press Enter at each question, then send us the line that the second command prints. It is your public key and safe to share.

ssh-keygen -t ed25519 -C "you@company.com"
cat ~/.ssh/id_ed25519.pub

Never share the file without “.pub”. That is your private key.

4

Connect to your server

Your welcome email has the server’s address and user name. Replace the example address below with yours.

ssh ubuntu@203.0.113.10
5

Check your GPUs

This lists every GPU on the server with its memory and current load. If you ordered 8 GPUs, you should see 8 rows.

nvidia-smi
6

Serve your first model

vLLM turns an open model into an OpenAI-compatible API. This example uses Qwen 2.5 7B, which runs on a single GPU and needs no sign-up.

python3 -m venv ~/venv && source ~/venv/bin/activate
pip install vllm
vllm serve Qwen/Qwen2.5-7B-Instruct

When the log shows the server is running, open a second terminal on the server and ask it a question:

curl http://localhost:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model": "Qwen/Qwen2.5-7B-Instruct", "messages": [{"role": "user", "content": "Hello from zaryacloud!"}]}'

Bigger model? Add --tensor-parallel-size 8 to spread it across 8 GPUs.

7

Open Jupyter in your browser

Start Jupyter on the server:

pip install jupyterlab
jupyter lab --no-browser --port 8888

Then, on your laptop, open a secure tunnel and visit http://localhost:8888. Use the token shown in the server’s log.

ssh -L 8888:localhost:8888 ubuntu@203.0.113.10
8

Move your data

Copy a folder from your laptop to the server. It resumes if the connection drops:

rsync -avP ./dataset/ ubuntu@203.0.113.10:~/data/

For object storage, use any S3 tool with the endpoint and keys from your welcome email:

aws s3 cp ./checkpoints s3://your-bucket/checkpoints --recursive \
  --endpoint-url https://YOUR-ENDPOINT-FROM-WELCOME-EMAIL