Implemented XLMRoberta sub-tasks

src/mlx_transformers/models/__init__.py

@@ -11,4 +11,10 @@
     RobertaForTokenClassification,
     RobertaForQuestionAnswering
 ) 
-from .xlm_roberta import XLMRobertaModel
+from .xlm_roberta import (
+    XLMRobertaModel,
+    XLMRobertaForSequenceClassification,
+    XLMRobertaForTokenClassification,
+    XLMRobertaForQuestionAnswering
+)
+

src/mlx_transformers/models/roberta.py

@@ -524,7 +524,9 @@ def __call__(
             if self.config.problem_type is None:
                 if self.num_labels == 1:
                     self.config.problem_type = "regression"
-                elif self.num_labels > 1 and (labels.dtype == mx.array):
+                elif self.num_labels > 1 and (
+                    labels.dtype == mx.long or labels.dtype == mx.int
+                    ):
                     self.config.problem_type = "single_label_classification"
                 else:
                     self.config.problem_type = "multi_label_classification"

src/mlx_transformers/models/xlm_roberta.py

@@ -451,6 +451,235 @@ def __call__(
         )
 
 
+class XLMRobertaClassificationHead(nn.Module):
+    """Head for sentence-level classification tasks."""
+
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        classifier_dropout = (
+            config.classifier_dropout if config.classifier_dropout is not None else config.hidden_dropout_prob
+        )
+        self.dropout = nn.Dropout(classifier_dropout)
+        self.out_proj = nn.Linear(config.hidden_size, config.num_labels)
+
+    def __call__(self, features, **kwargs):
+        x = features[:, 0, :]  # take <s> token (equiv. to [CLS])
+        x = self.dropout(x)
+        x = self.dense(x)
+        x = mx.tanh(x)
+        x = self.dropout(x)
+        x = self.out_proj(x)
+        return x
+
+
+
 class XLMRobertaForSequenceClassification(nn.Module):
     def __init__(self, config):
-        pass
+        super().__init__()
+        self.num_labels = config.num_labels
+        self.config = config
+
+        self.roberta = XLMRobertaModel(config, add_pooling_layer=False)
+        self.classifier = XLMRobertaClassificationHead(config)
+
+
+    def __call__(
+        self,
+        input_ids: Optional[mx.array] = None,
+        attention_mask: Optional[mx.array] = None,
+        token_type_ids: Optional[mx.array] = None,
+        position_ids: Optional[mx.array] = None,
+        labels: Optional[mx.array] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple[mx.array], SequenceClassifierOutput]:
+
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.roberta(
+            input_ids,
+            attention_mask=attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        sequence_output = outputs.last_hidden_state
+        logits = self.classifier(sequence_output)
+
+        loss = None
+        if labels is not None:
+            # move labels to correct device to enable model parallelism
+            labels = labels.to(logits.device)
+            if self.config.problem_type is None:
+                if self.num_labels == 1:
+                    self.config.problem_type = "regression"
+                elif self.num_labels > 1 and (
+                    labels.dtype == mx.long or labels.dtype == mx.int
+                    ):
+                    self.config.problem_type = "single_label_classification"
+                else:
+                    self.config.problem_type = "multi_label_classification"
+
+            if self.config.problem_type == "regression":
+                loss_fct = nn.losses.mse_loss()
+                if self.num_labels == 1:
+                    loss = loss_fct(logits.squeeze(), labels.squeeze())
+                else:
+                    loss = loss_fct(logits, labels)
+            elif self.config.problem_type == "single_label_classification":
+                loss_fct = nn.losses.cross_entropy()
+                loss = loss_fct(
+                    logits.view(-1, self.num_labels), labels.view(-1))
+            elif self.config.problem_type == "multi_label_classification":
+                loss_fct = nn.losses.binary_cross_entropy()
+                loss = loss_fct(logits, labels)
+
+        if not return_dict:
+            output = (logits,) + outputs[2:]
+            return ((loss,) + output) if loss is not None else output
+
+        return SequenceClassifierOutput(
+            loss=loss,
+            logits=logits,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+
+class XLMRobertaForTokenClassification(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.num_labels = config.num_labels
+        self.config = config
+
+        self.roberta = XLMRobertaModel(config, add_pooling_layer=False)
+        classifier_dropout = (
+            config.classifier_dropout if config.classifier_dropout is not None else config.hidden_dropout_prob
+        )
+        self.dropout = nn.Dropout(classifier_dropout)
+        self.classifier = nn.Linear(config.hidden_size, config.num_labels)
+
+
+    def __call__(
+        self,
+        input_ids: Optional[mx.array] = None,
+        attention_mask: Optional[mx.array] = None,
+        token_type_ids: Optional[mx.array] = None,
+        position_ids: Optional[mx.array] = None,
+        labels: Optional[mx.array] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple[mx.array], TokenClassifierOutput]:
+
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.roberta(
+            input_ids,
+            attention_mask=attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        sequence_output = outputs.last_hidden_state
+
+        sequence_output = self.dropout(sequence_output)
+        logits = self.classifier(sequence_output)
+
+        loss = None
+        if labels is not None:
+            # move labels to correct device to enable model parallelism
+            labels = labels.to(logits.device)
+            loss_fct = nn.losses.cross_entropy()
+            loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
+
+        if not return_dict:
+            output = (logits,) + outputs[2:]
+            return ((loss,) + output) if loss is not None else output
+
+        return TokenClassifierOutput(
+            loss=loss,
+            logits=logits,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+class XLMRobertaForQuestionAnswering(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.num_labels = config.num_labels
+        self.config = config 
+
+        self.roberta = XLMRobertaModel(config, add_pooling_layer=False)
+        self.qa_outputs = nn.Linear(config.hidden_size, config.num_labels)
+
+    def __call__(
+        self,
+        input_ids: Optional[mx.array] = None,
+        attention_mask: Optional[mx.array] = None,
+        token_type_ids: Optional[mx.array] = None,
+        position_ids: Optional[mx.array] = None,
+        start_positions: Optional[mx.array] = None,
+        end_positions: Optional[mx.array] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple[mx.array], QuestionAnsweringModelOutput]:
+
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.roberta(
+            input_ids,
+            attention_mask=attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        sequence_output = outputs.last_hidden_state
+
+        logits = self.qa_outputs(sequence_output)
+        splits = logits.split(2, axis=-1)
+        start_logits, end_logits = splits[0], splits[1]
+        start_logits = start_logits.squeeze(-1)
+        end_logits = end_logits.squeeze(-1)
+
+        total_loss = None
+        if start_positions is not None and end_positions is not None:
+            # If we are on multi-GPU, split add a dimension
+            if len(start_positions.size()) > 1:
+                start_positions = start_positions.squeeze(-1)
+            if len(end_positions.size()) > 1:
+                end_positions = end_positions.squeeze(-1)
+            # sometimes the start/end positions are outside our model inputs,
+            # we ignore these terms
+            ignored_index = start_logits.size(1)
+            start_positions = start_positions.clamp(0, ignored_index)
+            end_positions = end_positions.clamp(0, ignored_index)
+
+            loss_fct = nn.losses.cross_entropy()
+            start_loss = loss_fct(start_logits, start_positions)
+            end_loss = loss_fct(end_logits, end_positions)
+            total_loss = (start_loss + end_loss) / 2
+
+        if not return_dict:
+            output = (start_logits, end_logits) + outputs[2:]
+            return ((total_loss,) +
+                    output) if total_loss is not None else output
+
+        return QuestionAnsweringModelOutput(
+            loss=total_loss,
+            start_logits=start_logits,
+            end_logits=end_logits,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )