Module `neuralops.config.experiment_config`

Classes

class DatasetConfig (name: str, path: str, task_type: str, num_choices: int = 0, num_labels: int = 2)

Expand source code

@dataclass
class DatasetConfig:
    name: str
    path: str
    task_type: str  # "mc" or "single"
    num_choices: int = 0
    num_labels: int = 2

DatasetConfig(name: str, path: str, task_type: str, num_choices: int = 0, num_labels: int = 2)

Instance variables

var name : str: The type of the None singleton.
var num_choices : int: The type of the None singleton.
var num_labels : int: The type of the None singleton.
var path : str: The type of the None singleton.
var task_type : str: The type of the None singleton.

class ExperimentConfig (base_models: dict[str, str] = <factory>, svd_ranks: list[int] = <factory>, datasets: list[DatasetConfig] = <factory>, min_samples: int = 10, max_samples: int = 200, batch_size: int = 4, epochs: int = 1, learning_rate: float = 2e-05, max_length: int = 128, val_split: float = 0.2, seed: int = 42, gradient_accumulation_steps: int = 1, optimizer_type: str = 'adamw', lr_scheduler_type: str = 'linear', warmup_steps: int = 0, weight_decay: float = 0.01, mixed_precision: str = 'no', gradient_checkpointing: bool = False, max_grad_norm: float = 1.0, save_checkpoint_interval: int = 10, output_dir: str = './results', plots_dir: str = './plots/comprehensive', log_dir: str = './logs', selected_models: list[str] = <factory>, selected_datasets: list[str] = <factory>, selected_ranks: list[int] = <factory>)

Expand source code

@dataclass
class ExperimentConfig:
    base_models: dict[str, str] = field(default_factory=lambda: {"bert": "bert-base-uncased"})
    svd_ranks: list[int] = field(default_factory=lambda: [8, 16, 32])
    datasets: list[DatasetConfig] = field(default_factory=list)
    min_samples: int = 10
    max_samples: int = 200
    batch_size: int = 4
    epochs: int = 1
    learning_rate: float = 2e-5
    max_length: int = 128
    val_split: float = 0.2
    seed: int = 42
    gradient_accumulation_steps: int = 1

    # Advanced Training & Linear-Time Scalability Options
    optimizer_type: str = "adamw"  # Options: adamw, sgd, adafactor
    lr_scheduler_type: str = "linear"  # Options: linear, cosine, cosine_with_restarts
    warmup_steps: int = 0
    weight_decay: float = 0.01
    mixed_precision: str = "no"  # Options: no, fp16, bf16
    gradient_checkpointing: bool = (
        False  # Useful for extremely long sequences / linear time tracking
    )
    max_grad_norm: float = 1.0  # Gradient clipping threshold

    save_checkpoint_interval: int = 10
    output_dir: str = "./results"
    plots_dir: str = "./plots/comprehensive"
    log_dir: str = "./logs"
    selected_models: list[str] = field(default_factory=list)
    selected_datasets: list[str] = field(default_factory=list)
    selected_ranks: list[int] = field(default_factory=list)

ExperimentConfig(base_models: dict[str, str] = , svd_ranks: list[int] = , datasets: list[neuralops.config.experiment_config.DatasetConfig] = , min_samples: int = 10, max_samples: int = 200, batch_size: int = 4, epochs: int = 1, learning_rate: float = 2e-05, max_length: int = 128, val_split: float = 0.2, seed: int = 42, gradient_accumulation_steps: int = 1, optimizer_type: str = 'adamw', lr_scheduler_type: str = 'linear', warmup_steps: int = 0, weight_decay: float = 0.01, mixed_precision: str = 'no', gradient_checkpointing: bool = False, max_grad_norm: float = 1.0, save_checkpoint_interval: int = 10, output_dir: str = './results', plots_dir: str = './plots/comprehensive', log_dir: str = './logs', selected_models: list[str] = , selected_datasets: list[str] = , selected_ranks: list[int] = )

Instance variables

var base_models : dict[str, str]: The type of the None singleton.
var batch_size : int: The type of the None singleton.
var datasets : list[DatasetConfig]: The type of the None singleton.
var epochs : int: The type of the None singleton.
var gradient_accumulation_steps : int: The type of the None singleton.
var gradient_checkpointing : bool: The type of the None singleton.
var learning_rate : float: The type of the None singleton.
var log_dir : str: The type of the None singleton.
var lr_scheduler_type : str: The type of the None singleton.
var max_grad_norm : float: The type of the None singleton.
var max_length : int: The type of the None singleton.
var max_samples : int: The type of the None singleton.
var min_samples : int: The type of the None singleton.
var mixed_precision : str: The type of the None singleton.
var optimizer_type : str: The type of the None singleton.
var output_dir : str: The type of the None singleton.
var plots_dir : str: The type of the None singleton.
var save_checkpoint_interval : int: The type of the None singleton.
var seed : int: The type of the None singleton.
var selected_datasets : list[str]: The type of the None singleton.
var selected_models : list[str]: The type of the None singleton.
var selected_ranks : list[int]: The type of the None singleton.
var svd_ranks : list[int]: The type of the None singleton.
var val_split : float: The type of the None singleton.
var warmup_steps : int: The type of the None singleton.
var weight_decay : float: The type of the None singleton.