gaia.training.config module

Production Training Configuration System

class gaia.training.config.OptimizationType(value)

Bases: Enum

ADAM = 'adam'

ADAMW = 'adamw'

SGD = 'sgd'

RMSPROP = 'rmsprop'

LION = 'lion'

CATEGORICAL_ADAM = 'categorical_adam'

classmethod __contains__(member)

Return True if member is a member of this enum raises TypeError if member is not an enum member

note: in 3.12 TypeError will no longer be raised, and True will also be returned if member is the value of a member in this enum

classmethod __getitem__(name)

Return the member matching name.

classmethod __iter__()

Return members in definition order.

classmethod __len__()

Return the number of members (no aliases)

class gaia.training.config.SchedulerType(value)

Bases: Enum

COSINE = 'cosine'

LINEAR = 'linear'

EXPONENTIAL = 'exponential'

PLATEAU = 'plateau'

CATEGORICAL_ANNEALING = 'categorical_annealing'

classmethod __contains__(member)

Return True if member is a member of this enum raises TypeError if member is not an enum member

note: in 3.12 TypeError will no longer be raised, and True will also be returned if member is the value of a member in this enum

classmethod __getitem__(name)

Return the member matching name.

classmethod __iter__()

Return members in definition order.

classmethod __len__()

Return the number of members (no aliases)

class gaia.training.config.OptimizationConfig(optimizer=OptimizationType.ADAMW, learning_rate=0.001, weight_decay=0.0001, momentum=0.9, betas=(0.9, 0.999), eps=1e-08, categorical_coherence_weight=1.0, simplicial_regularization=0.1, horn_solving_weight=0.5, scheduler=SchedulerType.COSINE, scheduler_params=<factory>, warmup_steps=1000, gradient_clip_norm=1.0, gradient_accumulation_steps=1)

Bases: object

Optimization configuration with GAIA categorical extensions

optimizer: OptimizationType = 'adamw'

learning_rate: float = 0.001

weight_decay: float = 0.0001

momentum: float = 0.9

betas: tuple = (0.9, 0.999)

eps: float = 1e-08

categorical_coherence_weight: float = 1.0

simplicial_regularization: float = 0.1

horn_solving_weight: float = 0.5

scheduler: SchedulerType = 'cosine'

scheduler_params: Dict[str, Any]

warmup_steps: int = 1000

gradient_clip_norm: float | None = 1.0

gradient_accumulation_steps: int = 1

__init__(optimizer=OptimizationType.ADAMW, learning_rate=0.001, weight_decay=0.0001, momentum=0.9, betas=(0.9, 0.999), eps=1e-08, categorical_coherence_weight=1.0, simplicial_regularization=0.1, horn_solving_weight=0.5, scheduler=SchedulerType.COSINE, scheduler_params=<factory>, warmup_steps=1000, gradient_clip_norm=1.0, gradient_accumulation_steps=1)

class gaia.training.config.ModelConfig(name='gaia_model', architecture='categorical_mlp', hidden_dims=<factory>, activation='relu', dropout=0.1, batch_norm=True, simplicial_depth=3, categorical_embedding_dim=64, horn_solver_config=<factory>, init_method='xavier_uniform', init_gain=1.0)

Bases: object

Model architecture configuration

name: str = 'gaia_model'

architecture: str = 'categorical_mlp'

hidden_dims: List[int]

activation: str = 'relu'

dropout: float = 0.1

batch_norm: bool = True

simplicial_depth: int = 3

categorical_embedding_dim: int = 64

horn_solver_config: Dict[str, Any]

init_method: str = 'xavier_uniform'

init_gain: float = 1.0

__init__(name='gaia_model', architecture='categorical_mlp', hidden_dims=<factory>, activation='relu', dropout=0.1, batch_norm=True, simplicial_depth=3, categorical_embedding_dim=64, horn_solver_config=<factory>, init_method='xavier_uniform', init_gain=1.0)

class gaia.training.config.DataConfig(dataset_path='', batch_size=32, num_workers=0, pin_memory=False, persistent_workers=False, normalize=True, augmentation=False, augmentation_config=<factory>, validation_split=0.2, test_split=0.1, random_seed=42, categorical_features=<factory>, simplicial_structure=None)

Bases: object

Data loading and processing configuration

dataset_path: str | Path = ''

batch_size: int = 32

num_workers: int = 0

pin_memory: bool = False

persistent_workers: bool = False

normalize: bool = True

augmentation: bool = False

augmentation_config: Dict[str, Any]

validation_split: float = 0.2

test_split: float = 0.1

random_seed: int = 42

categorical_features: List[str]

simplicial_structure: Dict[str, Any] | None = None

__init__(dataset_path='', batch_size=32, num_workers=0, pin_memory=False, persistent_workers=False, normalize=True, augmentation=False, augmentation_config=<factory>, validation_split=0.2, test_split=0.1, random_seed=42, categorical_features=<factory>, simplicial_structure=None)

class gaia.training.config.TrainingConfig(epochs=100, max_steps=None, eval_frequency=1000, save_frequency=5000, log_level='INFO', log_frequency=100, use_tensorboard=True, use_wandb=False, wandb_project='gaia-training', checkpoint_dir='checkpoints', save_top_k=3, monitor_metric='val_loss', monitor_mode='min', early_stopping=True, patience=10, min_delta=0.0001, device='auto', mixed_precision=True, compile_model=True, distributed=False, world_size=1, rank=0, categorical_training=True, hierarchical_learning=True, horn_solving_frequency=1000, coherence_check_frequency=500, model=<factory>, data=<factory>, optimization=<factory>)

Bases: object

Complete training configuration

epochs: int = 100

max_steps: int | None = None

eval_frequency: int = 1000

save_frequency: int = 5000

log_level: str = 'INFO'

log_frequency: int = 100

use_tensorboard: bool = True

use_wandb: bool = False

wandb_project: str = 'gaia-training'

checkpoint_dir: str | Path = 'checkpoints'

save_top_k: int = 3

monitor_metric: str = 'val_loss'

monitor_mode: str = 'min'

early_stopping: bool = True

patience: int = 10

min_delta: float = 0.0001

device: str = 'auto'

mixed_precision: bool = True

compile_model: bool = True

distributed: bool = False

world_size: int = 1

rank: int = 0

categorical_training: bool = True

hierarchical_learning: bool = True

horn_solving_frequency: int = 1000

coherence_check_frequency: int = 500

model: ModelConfig

data: DataConfig

optimization: OptimizationConfig

classmethod from_file(config_path)

Load configuration from YAML or JSON file

save(config_path)

Save configuration to file

__init__(epochs=100, max_steps=None, eval_frequency=1000, save_frequency=5000, log_level='INFO', log_frequency=100, use_tensorboard=True, use_wandb=False, wandb_project='gaia-training', checkpoint_dir='checkpoints', save_top_k=3, monitor_metric='val_loss', monitor_mode='min', early_stopping=True, patience=10, min_delta=0.0001, device='auto', mixed_precision=True, compile_model=True, distributed=False, world_size=1, rank=0, categorical_training=True, hierarchical_learning=True, horn_solving_frequency=1000, coherence_check_frequency=500, model=<factory>, data=<factory>, optimization=<factory>)