Training Configs

Training uses one top-level dataclass plus nested optimizer and scheduler configs.

Public Imports

from artifex.configs import OptimizerConfig, SchedulerConfig, TrainingConfig

training_config = TrainingConfig(
    name="baseline_training",
    batch_size=64,
    num_epochs=100,
    optimizer=OptimizerConfig(
        name="adamw",
        optimizer_type="adamw",
        learning_rate=2e-4,
    ),
    scheduler=SchedulerConfig(
        name="cosine",
        scheduler_type="cosine",
        warmup_steps=500,
    ),
)

Available Types

• TrainingConfig
• OptimizerConfig
• SchedulerConfig

Key Training Fields

• batch_size
• num_epochs
• gradient_clip_norm
• checkpoint_dir
• save_frequency
• max_checkpoints
• log_frequency
• use_wandb
• wandb_project

The nested optimizer and scheduler configs hold the optimizer-specific and scheduler-specific parameters.