Skip to main content

​
Quick Start Guide

Get your first foundation model training in under an hour with these simple steps.
Two Ways to Work with Nolano AI: This guide shows both the Python SDK (pynolano) for creating configuration files and the CLI (nolano) for executing operations. Both work together seamlessly - you write configs with the SDK and run them with the CLI.
1

Prepare Your Data

Nolano.AI handles all data preprocessing with a single command. Here’s a minimal config example for text data:
Data Formats: Make sure your data follows the supported data formats for your specific modality.
data_config.py
from pynolano import DataPreparationConfig

def build() -> DataPreparationConfig:
    return DataPreparationConfig(
        input_path = "./raw_data", # Can also be a readable  stream
        output_path = "./prepared_data",
        tokenization = "Qwen/Qwen3-4B",
    )
Run data preparation:
nolano prepare_data data_config.py
2

Train Your Model

Create a simple training configuration:
train_config.py
from pynolano import DataConfig, OptimizationConfig, ExperimentConfig, ModelConfig

def build() -> ExperimentConfig:
    return ExperimentConfig(
        data_configs=DataConfig(data_paths="./prepared_data"),
        model_config=ModelConfig("Qwen/Qwen3-4B"),
        optimization_config=OptimizationConfig(
            global_batch_size=32, 
            max_learning_rate=3e-4, 
            total_training_steps=1000
        )
    )
Start training:
nolano train train_config.py
That’s it! Your model is now training with automatic:

Distributed Training

Data parallelism across available GPUs→ More details coming soon

Mixed Precision

Optimal performance with mixed precision training→ See precision parameter

Checkpointing

Checkpoint saving every epoch→ See MetaConfig parameters

Metrics Logging

Real-time metrics logging→ See WandbConfig parameters
3

Next Steps

After model training you can:
nolano convert_to_hf ./checkpoints/global_step_1000 config.yaml ./hf_model
Need help? reach out to support.