hydra.utils.model

PyTorch models for Hydra are defined here.

class hydra.utils.model.BertForMultiTargetTokenClassification(config: PreTrainedConfig, *args: Any, **kwargs: Any)[source]

Bases: MultiTargetTokenClassification, BertPreTrainedModel

A Hydra PyTorch model using a language model of BERT architecture.

config_class

alias of BertConfig

property lm_layer: Module
class hydra.utils.model.ModernBertForMultiTargetTokenClassification(config: ModernBertConfig, *args: Any, **kwargs: Any)[source]

Bases: MultiTargetTokenClassification, ModernBertPreTrainedModel

A Hydra PyTorch model using a language model of ModernBert architecture.

config_class

alias of ModernBertConfig

property lm_layer: Module
process_lm_output(sequence_output: Tensor) → Tensor[source]
class hydra.utils.model.MultiTargetTokenClassification(config: PreTrainedConfig, **kwargs: Any)[source]

Bases: PreTrainedModel

Base class for Hydra PyTorch models.

forward(input_ids: Tensor | None = None, attention_mask: Tensor | None = None, token_type_ids: Tensor | None = None, position_ids: Tensor | None = None, inputs_embeds: Tensor | None = None, labels: Tensor | None = None, **kwargs: Any) → tuple[Tensor] | TokenClassifierOutput[source]

Run a forward pass of the model.

Returns:

model outputs.

linear_size = 128
property lm_layer: Module

The underlying language model.

Subclasses must implement this method.

Raises:

NotImplementedError – Always.

lm_layer_input_keys = ('input_ids', 'attention_mask', 'token_type_ids', 'position_ids', 'head_mask', 'inputs_embeds', 'output_attentions', 'output_hidden_states', 'return_dict', 'training')
process_lm_output(sequence_output: Tensor) → Tensor[source]

Perform any additional operations on the underlying language model’s output.

Override in subclasses for specific behaviour.

Returns:

its input.

set_loss_weights(loss_weights: dict[str, float]) → None[source]

Assign weights to losses from particular model outputs.

class hydra.utils.model.RobertaForMultiTargetTokenClassification(config: PreTrainedConfig, *args: Any, **kwargs: Any)[source]

Bases: MultiTargetTokenClassification, RobertaPreTrainedModel

A Hydra PyTorch model using a lanugage model of RoBERTa architecture.

config_class

alias of RobertaConfig

property lm_layer: Module
class hydra.utils.model.T5ForMultiTargetTokenClassification(config: T5Config, *args: Any, **kwargs: Any)[source]

Bases: MultiTargetTokenClassification, T5PreTrainedModel

A Hydra PyTorch model using a language model of T5 architecture.

config_class

alias of T5Config

property lm_layer: Module
class hydra.utils.model.XLMRobertaForMultiTargetTokenClassification(config: PreTrainedConfig, *args: Any, **kwargs: Any)[source]

Bases: MultiTargetTokenClassification, XLMRobertaPreTrainedModel

A Hydra PyTorch model using a language model of XMLRoBERTa architecture.

config_class

alias of XLMRobertaConfig

property lm_layer: Module
hydra.utils.model.get_model_class(model_type: str) → type[MultiTargetTokenClassification][source]

Provide a class for Hydra PyTorch model based on language model type.

Returns:

the class.

Raises:

KeyError – for unsupported model types.

hydra.utils.model.get_model_type(pretrained_path: str) → str[source]

Retrieve the type of pretrained language model from its config.

Returns:

model type identifier (e.g. ‘bert’).