Put relative import first for Huggingface

modeling_chatglm.py

@@ -24,15 +24,17 @@ from transformers.generation.utils import GenerationMixin
 
 
 try:
-    from configuration_chatglm import ChatGLMConfig
-    from ops.pooling import mean_pooling
-    from ops.compressed_attention import compressed_attention
-    from ops.topk_sparse_attention import topk_sparse_attention
-except ImportError:
     from .configuration_chatglm import ChatGLMConfig
     from .ops.pooling import mean_pooling
     from .ops.compressed_attention import compressed_attention
     from .ops.topk_sparse_attention import topk_sparse_attention
+except ImportError:
+    from configuration_chatglm import ChatGLMConfig
+    from ops.pooling import mean_pooling
+    from ops.compressed_attention import compressed_attention
+    from ops.topk_sparse_attention import topk_sparse_attention
+
+
 
 try:
     from transformers.utils import is_flash_attn_greater_or_equal_2_10, is_flash_attn_2_available