Update modeling_hyperclovax.py (#6)

- Update modeling_hyperclovax.py (1f212e3e2f00d627de3161750980468df377dc1d)

modeling_hyperclovax.py

@@ -1,6 +1,7 @@
 # coding=utf-8
 # This file was created for the HyperCLOVA X SEED 14B Think architecture.
-# partially copied and modified from https://github.com/huggingface/transformers
+# partially copied and modified from
+# https://github.com/huggingface/transformers/blob/v4.52.4/src/transformers/models/llama/modeling_llama.py
 # Copyright 2022 EleutherAI and the HuggingFace Inc. team. All rights reserved.
 #
 # This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
@@ -43,7 +44,14 @@ from transformers.modeling_rope_utils import ROPE_INIT_FUNCTIONS, dynamic_rope_u
 from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
 from transformers.processing_utils import Unpack
 from transformers.pytorch_utils import ALL_LAYERNORM_LAYERS
-from transformers.utils import LossKwargs, auto_docstring, can_return_tuple, is_torch_flex_attn_available, logging
+from transformers.utils import auto_docstring, can_return_tuple, is_torch_flex_attn_available, logging
+try:
+    from transformers.utils import LossKwargs
+    loss_kwargs_class = LossKwargs
+except ImportError:
+    from transformers.utils import TransformersKwargs
+    loss_kwargs_class = TransformersKwargs
+
 from .configuration_hyperclovax import HyperCLOVAXConfig
 if is_torch_flex_attn_available():
     from torch.nn.attention.flex_attention import BlockMask
@@ -620,7 +628,7 @@ class HyperCLOVAXModel(HyperCLOVAXPreTrainedModel):
         return causal_mask
 
 
-class KwargsForCausalLM(FlashAttentionKwargs, LossKwargs): ...
+class KwargsForCausalLM(FlashAttentionKwargs, loss_kwargs_class): ...
 
 
 @auto_docstring