update model

modeling_flexbert.py

@@ -1507,7 +1507,8 @@ class FlexBertForMultipleChoice(FlexBertPreTrainedModel):
 
 
 class FlexBertForCasualLM(FlexBertPreTrainedModel):
-    config_class = FlexBertConfig  # Add this line
+    config_class = FlexBertConfig 
+    
     """Bert Model transformer with a LM head.
 
     This head is just a standard LM head module. Used for causal language modeling tasks.
@@ -1701,8 +1702,8 @@ class FlexBertForCasualLM(FlexBertPreTrainedModel):
                 shift_labels.view(-1)
             )
 
-        assert False
         if self.pad_logits:
+            print(f"Padding logits: {logits.shape}")
             new_logits = self.pad_inputs(logits, indices, batch_size, seq_len)[0]
             if len(new_logits.shape) == 2:
                 new_logits = new_logits.unsqueeze(0)
@@ -1713,6 +1714,7 @@ class FlexBertForCasualLM(FlexBertPreTrainedModel):
                 attentions=None,
             )
         else:
+            print(f"Non-padding logits: {logits.shape}")
             if len(logits.shape) == 2:
                 logits = logits.unsqueeze(0)
             return CausalLMOutput(

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:96348b6ba9cd41656884fe36b14263dbb74f76da5632fa0f4b92c0781353f2b0
+oid sha256:67a284058719cc19a4ec63c15f1f3cd8b57e5d4e59494b1704b7ae33ab717fb9
 size 598685038