hezar.models.sequence_labeling.distilbert.distilbert_sequence_labeling module¶
A DISTILBERT model for sequence labeling built using HuggingFace Transformers
- class hezar.models.sequence_labeling.distilbert.distilbert_sequence_labeling.DistilBertSequenceLabeling(config: DistilBertSequenceLabelingConfig, **kwargs)[source]¶
Bases:
Model- compute_loss(logits: Tensor, labels: Tensor) Tensor[source]¶
Compute loss on the model outputs against the given labels
- Parameters:
logits – Logits tensor to compute loss on
labels – Labels tensor
Note: Subclasses can also override this method and add other arguments besides logits and labels
- Returns:
Loss tensor
- forward(token_ids, attention_mask=None, head_mask=None, inputs_embeds=None, output_attentions=None, output_hidden_states=None, **kwargs) Dict[source]¶
Forward inputs through the model and return logits, etc.
- Parameters:
model_inputs – The required inputs for the model forward
- Returns:
A dict of outputs like logits, loss, etc.
- post_process(model_outputs: Dict[str, Tensor], return_offsets: bool = False, return_scores: bool = False)[source]¶
Process model outputs and return human-readable results. Called in self.predict()
- Parameters:
model_outputs – model outputs to process
**kwargs – extra arguments specific to the derived class
- Returns:
Processed model output values and converted to human-readable results
- preprocess(inputs: str | List[str], **kwargs)[source]¶
Given raw inputs, preprocess the inputs and prepare them for model’s forward().
- Parameters:
raw_inputs – Raw model inputs
**kwargs – Extra kwargs specific to the model. See the model’s specific class for more info
- Returns:
A dict of inputs for model forward