rlightning.policy.utils.utils¶

rlightning.policy.utils.utils.expand_to_target_dim(tensor, target_shape)[source]¶

rlightning.policy.utils.utils.huber_loss(error: MockModule('torch.Tensor'), delta: float)[source]¶

rlightning.policy.utils.utils.masked_mean(values: MockModule('torch.Tensor'), mask: MockModule('torch.Tensor'), axis=None)[source]¶: Compute mean of tensor with a masked values.

rlightning.policy.utils.utils.masked_mean_ratio(values: MockModule('torch.Tensor'), mask: MockModule('torch.Tensor'), loss_mask_ratio: MockModule('torch.Tensor'))[source]¶

rlightning.policy.utils.utils.masked_sum(values: MockModule('torch.Tensor'), mask: MockModule('torch.Tensor'), axis=None)[source]¶: Compute mean of tensor with a masked values.

rlightning.policy.utils.utils.postprocess_loss_metric(metrics_data: dict) → dict[source]¶

rlightning.policy.utils.utils.preprocess_loss_inputs(logprobs: MockModule('torch.Tensor'), old_logprobs: MockModule('torch.Tensor'), advantages: MockModule('torch.Tensor'), logprob_type: str | None = None, single_action_dim: int | None = None, loss_mask: MockModule('torch.Tensor') | None = None, loss_mask_sum: MockModule('torch.Tensor') | None = None, values: MockModule('torch.Tensor') | None = None, prev_values: MockModule('torch.Tensor') | None = None, returns: MockModule('torch.Tensor') | None = None, reward_type: str | None = None, **kwargs) → dict[source]¶

rlightning.policy.utils.utils.seq_mean_token_mean(values: MockModule('torch.Tensor'), mask: MockModule('torch.Tensor'), dim: int = -1)[source]¶

rlightning.policy.utils.utils.seq_mean_token_sum(values: MockModule('torch.Tensor'), mask: MockModule('torch.Tensor'), dim: int = -1)[source]¶