Skip to content

Quick Start Guide

This guide will walk you through the essential steps to get Optimus-DL up and running, from installation to starting your first training job.

1. Installation

First, clone the repository and install the framework in editable mode. This allows you to easily modify the code and have your changes reflected immediately.

# Clone the repository
git clone https://github.com/alexdremov/optimus-dl
cd optimus-dl

# Install in editable mode with dependencies
pip install -e .

Alternatively, you can use docker: https://hub.docker.com/repository/docker/alexdremov/optimus-dl/general

2. Running a Training Job

The easiest way to start is to run a training job with one of the provided default configurations. The train_llama.yaml config is a good starting point.

Training is orchestrated by the scripts/train.py script.

# Run with the default Llama configuration
python scripts/train.py --config-name=train_llama

# multi-gpu training
torchrun --nproc_per_node=gpu scripts/train.py --config-name=train_llama

This command will: 1. Load the train_llama.yaml configuration. 2. Build the model, data pipeline, optimizer, and other components. 3. Start the training loop, which will log progress and save checkpoints to the outputs/ directory.

3. Customizing Your Run

You can easily override any parameter from the configuration file directly on the command line. This is perfect for quick experiments.

# Override the batch size and use a different model
python scripts/train.py \
    --config-name=train_llama \
    model=gpt2 \
    args.batch_size=32

This will start a new training run using the gpt2 model configuration and a batch size of 32, while keeping all other settings from train_llama.yaml.

4. Evaluation

Once you have a trained checkpoint, you can evaluate it on standard benchmarks using the integrated Language Model Evaluation Harness.

# Evaluate a checkpoint on Hellaswag and MMLU
python scripts/eval.py \
    common.checkpoint_path=outputs/my-run/checkpoint_00010000 \
    lm_eval.tasks=[hellaswag,mmlu] \
    lm_eval.batch_size=8

5. Serving the Model

Optimus-DL also includes a simple server to deploy your trained models as an OpenAI-compatible API endpoint.

# Serve a pre-configured TinyLlama model
python scripts/serve.py --config-name=tinyllama

You can then send requests to the running server: 

curl -X POST http://localhost:8000/v1/completions \
  -d '{"prompt": "The future of AI is", "max_tokens": 50}'

What's Next?

You've successfully run your first training job! Here's where to go next: