Added attn_imp in config

configuration_chatglm.py

@@ -17,6 +17,7 @@ class ChatGLMConfig(PretrainedConfig):
             kernel_size=64,
             kernel_stride=64,
             window_size=512,
+            attn_implementation="nsa",
             topk=16,
             init_blocks=1,
             local_blocks=2,
@@ -38,6 +39,7 @@ class ChatGLMConfig(PretrainedConfig):
             fp32_residual_connection=False,
             **kwargs
     ):
+        self.attn_implementation = attn_implementation
         self.num_layers = num_layers
         self.vocab_size = padded_vocab_size
         self.padded_vocab_size = padded_vocab_size