Logging Callbacks

Module: generative_models.training.callbacks.logging

Source: generative_models/training/callbacks/logging.py

Overview

Logging callbacks for integrating experiment tracking and progress monitoring into training loops. These callbacks wrap existing logging infrastructure for seamless integration with the callback system.

Classes

LoggerCallbackConfig

@dataclass(slots=True)
class LoggerCallbackConfig:
    log_every_n_steps: int = 1
    log_on_epoch_end: bool = True
    prefix: str = ""

Configuration for base logger callback.

Attributes:

• log_every_n_steps: Log metrics every N training steps
• log_on_epoch_end: Whether to log metrics at end of each epoch
• prefix: Prefix to add to metric names

---

LoggerCallback

class LoggerCallback(BaseCallback):
    def __init__(
        self,
        logger: Logger,
        config: Optional[LoggerCallbackConfig] = None,
    )

Base callback that wraps any Logger instance for training integration.

Parameters:

• logger: Logger instance to use for logging (from utils.logging)
• config: Logging configuration. Uses defaults if not provided

Example:

from artifex.generative_models.utils.logging import ConsoleLogger
from artifex.generative_models.training.callbacks import LoggerCallback

logger = ConsoleLogger(name="training")
callback = LoggerCallback(logger=logger)
trainer.fit(callbacks=[callback])

---

WandbLoggerConfig

@dataclass(slots=True)
class WandbLoggerConfig:
    project: str
    entity: Optional[str] = None
    name: Optional[str] = None
    tags: list[str] = field(default_factory=list)
    notes: Optional[str] = None
    config: dict[str, Any] = field(default_factory=dict)
    mode: Literal["online", "offline", "disabled"] = "online"
    resume: Literal["allow", "never", "must", "auto"] | bool | None = None
    log_every_n_steps: int = 1
    log_on_epoch_end: bool = True
    log_dir: Optional[str] = None

Configuration for W&B logging callback.

Attributes:

• project: W&B project name (required)
• entity: W&B entity (username or team name)
• name: Run name. If None, W&B auto-generates
• tags: List of tags for the run
• notes: Notes about the run
• config: Dictionary of hyperparameters to log
• mode: W&B mode: "online", "offline", or "disabled"
• resume: Whether to resume a previous run
• log_every_n_steps: Log metrics every N training steps
• log_on_epoch_end: Whether to log metrics at end of each epoch
• log_dir: Local directory for W&B files

---

WandbLoggerCallback

class WandbLoggerCallback(BaseCallback):
    def __init__(self, config: WandbLoggerConfig)

Weights & Biases experiment tracking callback.

Features:

• Automatic metric logging
• Hyperparameter tracking
• Run resumption support
• Multiple run modes (online, offline, disabled)

Example:

from artifex.generative_models.training.callbacks import (
    WandbLoggerCallback,
    WandbLoggerConfig,
)

config = WandbLoggerConfig(
    project="my-project",
    name="experiment-1",
    tags=["baseline", "vae"],
    config={"learning_rate": 1e-3, "batch_size": 32},
)
callback = WandbLoggerCallback(config=config)
trainer.fit(callbacks=[callback])

---

TensorBoardLoggerConfig

@dataclass(slots=True)
class TensorBoardLoggerConfig:
    log_dir: str = "logs/tensorboard"
    flush_secs: int = 120
    max_queue: int = 10
    log_every_n_steps: int = 1
    log_on_epoch_end: bool = True
    log_graph: bool = False

Configuration for TensorBoard logging callback.

Attributes:

• log_dir: Directory for TensorBoard logs
• flush_secs: How often to flush to disk (seconds)
• max_queue: Max queue size for pending events
• log_every_n_steps: Log metrics every N training steps
• log_on_epoch_end: Whether to log metrics at end of each epoch
• log_graph: Whether to log model graph

---

TensorBoardLoggerCallback

class TensorBoardLoggerCallback(BaseCallback):
    def __init__(self, config: TensorBoardLoggerConfig)

TensorBoard logging callback.

Features:

• Scalar metrics logging
• Configurable flush interval
• Per-step and per-epoch logging

Requirements: Requires tensorboard package (pip install tensorboard)

Example:

from artifex.generative_models.training.callbacks import (
    TensorBoardLoggerCallback,
    TensorBoardLoggerConfig,
)

config = TensorBoardLoggerConfig(
    log_dir="logs/experiment-1",
    flush_secs=60,
)
callback = TensorBoardLoggerCallback(config=config)
trainer.fit(callbacks=[callback])

---

ProgressBarConfig

@dataclass(slots=True)
class ProgressBarConfig:
    refresh_rate: int = 10
    show_eta: bool = True
    show_metrics: bool = True
    leave: bool = True
    disable: bool = False

Configuration for progress bar callback.

Attributes:

• refresh_rate: How often to refresh the progress bar (steps)
• show_eta: Whether to show estimated time of arrival
• show_metrics: Whether to display metrics in progress bar
• leave: Whether to leave progress bar after completion
• disable: Whether to disable progress bar entirely

---

ProgressBarCallback

class ProgressBarCallback(BaseCallback):
    def __init__(self, config: Optional[ProgressBarConfig] = None)

Rich console progress bar callback.

Features:

• Real-time training progress
• Metric display
• ETA estimation
• Nested progress for epochs/steps

Requirements: Requires rich package (pip install rich)

Example:

from artifex.generative_models.training.callbacks import (
    ProgressBarCallback,
    ProgressBarConfig,
)

config = ProgressBarConfig(
    refresh_rate=10,
    show_eta=True,
    show_metrics=True,
)
callback = ProgressBarCallback(config=config)
trainer.fit(callbacks=[callback])

Usage with Multiple Callbacks

Logging callbacks can be combined with other callbacks:

from artifex.generative_models.training.callbacks import (
    CallbackList,
    EarlyStopping,
    EarlyStoppingConfig,
    ModelCheckpoint,
    CheckpointConfig,
    WandbLoggerCallback,
    WandbLoggerConfig,
    ProgressBarCallback,
)

# Configure callbacks
callbacks = CallbackList([
    # Logging
    WandbLoggerCallback(WandbLoggerConfig(
        project="my-project",
        name="experiment-1",
    )),
    ProgressBarCallback(),

    # Training control
    EarlyStopping(EarlyStoppingConfig(
        monitor="val_loss",
        patience=10,
    )),

    # Checkpointing
    ModelCheckpoint(CheckpointConfig(
        dirpath="checkpoints",
        monitor="val_loss",
    )),
])

trainer.fit(callbacks=callbacks)

Module Statistics

• Classes: 8
• Functions: 0
• Imports: 6