Quickstart

​

Installation

You can install VESSL CLI through pip.

pip install --upgrade vessl

​

Train nanoGPT with VESSL Run

To help you get started, we prepared a quickstart command that holds several example YAML files for popular open-source models on GitHub. The following command prompts a list of example models. At this step, you will be asked to log in and grant access permission.

vessl hello

Select nanogpt from a list of models. This initiates a VESSL Run with the following nanogpt.yaml file, which you can also check on your terminal as the Run starts.

name: nanogpt
image: nvcr.io/nvidia/pytorch:22.03-py3
resources:
  cluster: aws-apne2
  preset: v1.v100-1.mem-52
import:
  /root/examples: git://github.com/vessl-ai/examples
export:
  /output: vessl-artifact://
run:
  - workdir: /root/examples/nanogpt
    command: |
      pip install torchaudio -f https://download.pytorch.org/whl/cu111/torch_stable.html
      pip install transformers datasets tiktoken wandb tqdm
      python data/shakespeare_char/prepare.py
      python train.py config/train_shakespeare_char.py
      python sample.py --out_dir=out-shakespeare-char

The command performs the following as defined in the YAML file:

1. Launch a training job & cluster on AWS with 1 NVIDIA V100 GPU.
2. Configure runtime with CUDA compute-capable PyTorch 22.03.
3. Mount the nanoGPT GitHub repo and set the working directory.
4. Run the task’s run commands defined under command.
5. Track training progress on VESSL.

Click the output link in your terminal to check the training progress for the Run along with the key metrics and hyperparameters.

You can also launch the same Run by copying and pasting the YAML above and running the following command.

vessl run -f nanogpt.yaml

​

What’s next

Run’s unified YAML interface really shines as you (1) fine-tune a model with your dataset, (2) scale it on your cloud or on-prems, (3) and create a micro AI/ML app. Follow the guides below to experiment with popular models like Dreambooth Stable Diffusion and Segment Anything using VESSL Run.