Add NeoBERTForTokenClassification class

config.json

@@ -6,7 +6,8 @@
     "AutoConfig": "model.NeoBERTConfig",
     "AutoModel": "model.NeoBERT",
     "AutoModelForMaskedLM": "model.NeoBERTLMHead",
-    "AutoModelForSequenceClassification": "model.NeoBERTForSequenceClassification"
+    "AutoModelForSequenceClassification": "model.NeoBERTForSequenceClassification",
+    "AutoModelForTokenClassification": "model.NeoBERTForTokenClassification"
   },
   "classifier_init_range": 0.02,
   "decoder_init_range": 0.02,
@@ -15,8 +16,28 @@
   "hidden_size": 768,
   "intermediate_size": 3072,
   "kwargs": {
+    "_commit_hash": null,
+    "architectures": [
+      "NeoBERTLMHead"
+    ],
+    "attn_implementation": null,
+    "auto_map": {
+      "AutoConfig": "model.NeoBERTConfig",
+      "AutoModel": "model.NeoBERT",
+      "AutoModelForMaskedLM": "model.NeoBERTLMHead",
+      "AutoModelForSequenceClassification": "model.NeoBERTForSequenceClassification"
+    },
     "classifier_init_range": 0.02,
+    "dim_head": 64,
+    "kwargs": {
+      "classifier_init_range": 0.02,
+      "pretrained_model_name_or_path": "google-bert/bert-base-uncased",
+      "trust_remote_code": true
+    },
+    "model_type": "neobert",
     "pretrained_model_name_or_path": "google-bert/bert-base-uncased",
+    "torch_dtype": "float32",
+    "transformers_version": "4.48.2",
     "trust_remote_code": true
   },
   "max_length": 4096,
@@ -27,7 +48,7 @@
   "pad_token_id": 0,
   "pretrained_model_name_or_path": "google-bert/bert-base-uncased",
   "torch_dtype": "float32",
-  "transformers_version": "4.48.2",
+  "transformers_version": "4.51.3",
   "trust_remote_code": true,
   "vocab_size": 30522
 }

model.py

@@ -27,6 +27,7 @@ from transformers.modeling_outputs import (
     BaseModelOutput,
     MaskedLMOutput,
     SequenceClassifierOutput,
+    TokenClassifierOutput
 )
 
 from .rotary import precompute_freqs_cis, apply_rotary_emb
@@ -432,3 +433,72 @@ class NeoBERTForSequenceClassification(NeoBERTPreTrainedModel):
             hidden_states=output.hidden_states if output_hidden_states else None,
             attentions=output.attentions if output_attentions else None,
         )
+
+class NeoBERTForTokenClassification(NeoBERTPreTrainedModel):
+    config_class = NeoBERTConfig
+
+    def __init__(self, config: NeoBERTConfig):
+        super().__init__(config)
+
+        self.config = config
+        self.num_labels = getattr(config, "num_labels", 2)
+        self.classifier_dropout = getattr(config, "classifier_dropout", 0.1)
+        self.classifier_init_range = getattr(config, "classifier_init_range", 0.02)
+
+        self.model = NeoBERT(config)
+
+        self.dropout = nn.Dropout(self.classifier_dropout)
+        self.classifier = nn.Linear(config.hidden_size, self.num_labels)
+
+        self.post_init()
+
+    def _init_weights(self, module):
+        if isinstance(module, nn.Linear):
+            module.weight.data.normal_(mean=0.0, std=self.classifier_init_range)
+            if module.bias is not None:
+                module.bias.data.zero_()
+
+    def forward(
+        self,
+        input_ids: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.Tensor] = None,
+        max_seqlen: Optional[int] = None,
+        cu_seqlens: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        output_hidden_states: Optional[bool] = False,
+        output_attentions: Optional[bool] = False,
+        labels: Optional[torch.Tensor] = None,
+        return_dict: Optional[bool] = None,
+    ):
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        output = self.model(
+            input_ids=input_ids,
+            position_ids=position_ids,
+            max_seqlen=max_seqlen,
+            cu_seqlens=cu_seqlens,
+            attention_mask=attention_mask,
+            output_hidden_states=output_hidden_states,
+            output_attentions=output_attentions,
+        )
+
+        sequence_output = output.last_hidden_state
+        sequence_output = self.dropout(sequence_output)
+        logits = self.classifier(sequence_output)
+
+        loss = None
+        if labels is not None:
+            loss_fct = CrossEntropyLoss()
+            # Reshape logits and labels to compute token classification loss
+            loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
+
+        if not return_dict:
+            output = (logits,)
+            return ((loss,) + output) if loss is not None else output
+
+        return TokenClassifierOutput(
+            loss=loss,
+            logits=logits,
+            hidden_states=output.hidden_states if output_hidden_states else None,
+            attentions=output.attentions if output_attentions else None,
+        )