opentau.policies.value.configuration_value

Configuration module for the Value policy.

This module defines the ValueConfig class, which handles the configuration parameters for the Value policy. It includes settings for the model architecture, optimization, scheduling, and data processing.

Classes

ValueConfig(n_obs_steps, ...)

Configuration class for the Value policy.

class opentau.policies.value.configuration_value.ValueConfig(n_obs_steps: int = 1, normalization_mapping: dict[str, ~opentau.configs.types.NormalizationMode] = <factory>, input_features: dict[str, ~opentau.configs.types.PolicyFeature] = <factory>, output_features: dict[str, ~opentau.configs.types.PolicyFeature] = <factory>, device: str | None = None, use_amp: bool = False, use_torch_compile: bool = False, torch_compile_mode: str = 'default', pretrained_path: str | None = None, skip_normalization_weights: bool = False, skip_input_resolution_check: bool = False, save_normalization_stats: bool = True, dataset_names: list[str] | None = None, dataset_to_norm_index: dict[str, int] | None = None, cloud_vlm_latency_mean: float = 0.0, cloud_vlm_latency_std: float = 0.0, cloud_vlm_latency_lower: float = 0.0, cloud_vlm_latency_upper: float = 0.0, action_decoder_latency_mean: float = 0.0, action_decoder_latency_std: float = 0.0, action_decoder_latency_lower: float = 0.0, action_decoder_latency_upper: float = 0.0, chunk_size: int = 50, max_state_dim: int = 32, resize_imgs_with_padding: tuple[int, int] = (224, 224), empty_cameras: int = 0, prompt_max_length: int = 48, response_max_length: int = 52, reward_config: ~opentau.configs.reward.RewardConfig = <factory>, optimizer_lr: float = 2.5e-05, optimizer_betas: tuple[float, float] = (0.9, 0.95), optimizer_eps: float = 1e-08, optimizer_weight_decay: float = 1e-10, scheduler_warmup_steps: int = 1000, scheduler_decay_steps: int = 30000, scheduler_decay_lr: float = 2.5e-06)

Bases: PreTrainedConfig

Configuration class for the Value policy.

Parameters:

• n_obs_steps – Number of observation steps to be used.

• chunk_size – The chunk size for the policy.

• normalization_mapping – Mapping of feature types to normalization modes.

• max_state_dim – Maximum dimension for state vectors.

• resize_imgs_with_padding – Tuple indicating the size to resize images with padding.

• empty_cameras – Number of empty cameras to add.

• tokenizer_max_length – Maximum length for the tokenizer.

• reward_config – Configuration for the reward.

• optimizer_lr – Learning rate for the optimizer.

• optimizer_betas – Betas for the optimizer.

• optimizer_eps – Epsilon for the optimizer.

• optimizer_weight_decay – Weight decay for the optimizer.

• scheduler_warmup_steps – Number of warmup steps for the scheduler.

• scheduler_decay_steps – Number of decay steps for the scheduler.

• scheduler_decay_lr – Decay learning rate for the scheduler.

__init__(n_obs_steps: int = 1, normalization_mapping: dict[str, ~opentau.configs.types.NormalizationMode] = <factory>, input_features: dict[str, ~opentau.configs.types.PolicyFeature] = <factory>, output_features: dict[str, ~opentau.configs.types.PolicyFeature] = <factory>, device: str | None = None, use_amp: bool = False, use_torch_compile: bool = False, torch_compile_mode: str = 'default', pretrained_path: str | None = None, skip_normalization_weights: bool = False, skip_input_resolution_check: bool = False, save_normalization_stats: bool = True, dataset_names: list[str] | None = None, dataset_to_norm_index: dict[str, int] | None = None, cloud_vlm_latency_mean: float = 0.0, cloud_vlm_latency_std: float = 0.0, cloud_vlm_latency_lower: float = 0.0, cloud_vlm_latency_upper: float = 0.0, action_decoder_latency_mean: float = 0.0, action_decoder_latency_std: float = 0.0, action_decoder_latency_lower: float = 0.0, action_decoder_latency_upper: float = 0.0, chunk_size: int = 50, max_state_dim: int = 32, resize_imgs_with_padding: tuple[int, int] = (224, 224), empty_cameras: int = 0, prompt_max_length: int = 48, response_max_length: int = 52, reward_config: ~opentau.configs.reward.RewardConfig = <factory>, optimizer_lr: float = 2.5e-05, optimizer_betas: tuple[float, float] = (0.9, 0.95), optimizer_eps: float = 1e-08, optimizer_weight_decay: float = 1e-10, scheduler_warmup_steps: int = 1000, scheduler_decay_steps: int = 30000, scheduler_decay_lr: float = 2.5e-06) → None

property action_delta_indices: list

Returns the action delta indices.

Returns:

List of indices from 0 to chunk_size.

Return type:

list

chunk_size: int = 50

empty_cameras: int = 0

get_optimizer_preset() → AdamWConfig

Returns the optimizer preset configuration.

Returns:

The optimizer configuration.

Return type:

AdamWConfig

get_scheduler_preset() → LRSchedulerConfig

Returns the scheduler preset configuration.

Returns:

The scheduler configuration.

Return type:

CosineDecayWithWarmupSchedulerConfig

max_state_dim: int = 32

n_obs_steps: int = 1

normalization_mapping: dict[str, NormalizationMode]

property observation_delta_indices: None

Returns the observation delta indices.

Returns:

Always returns None.

Return type:

None

optimizer_betas: tuple[float, float] = (0.9, 0.95)

optimizer_eps: float = 1e-08

optimizer_lr: float = 2.5e-05

optimizer_weight_decay: float = 1e-10

prompt_max_length: int = 48

resize_imgs_with_padding: tuple[int, int] = (224, 224)

response_max_length: int = 52

reward_config: RewardConfig

property reward_delta_indices: None

Returns the reward delta indices.

Returns:

Always returns None.

Return type:

None

scheduler_decay_lr: float = 2.5e-06

scheduler_decay_steps: int = 30000

scheduler_warmup_steps: int = 1000

validate_features() → None

Validates features and adds empty cameras if specified.