update model

modeling_flexbert.py

@@ -1642,7 +1642,7 @@ class FlexBertForCausalLM(FlexBertPreTrainedModel):
         if self.unpad_embeddings and (indices is None and cu_seqlens is None and max_seqlen is None):
             batch_size, seq_len = input_ids.shape[:2]
             if attention_mask is None:
-                # create all ones
+                # create all ones, except for padding (TODO?)
                 attention_mask = torch.ones_like(input_ids)
             input_ids, indices, cu_seqlens, max_seqlen, position_ids, labels = self.unpad_inputs(
                 input_ids, attention_mask, position_ids, labels

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:48c87e6a8828cf590bb5296043df2cf02dec98ec3c196baa44e0e7ae767f6455
+oid sha256:7863cc4c58494c661ffd3c77af90796f5caa8217917c2f6e7c99cc28d65b58c2
 size 598685038