fix the spell bug and move the sequence judge to training_internlm

internlm/train/training_internlm.py

@@ -113,6 +113,7 @@ def initialize_model():
     model = wrap_FSDP_model(model)
 
     # check whether the norm module has IS_SEQUENCE_PARALLEL attribute
+    if gpc.config.parallel.sequence_parallel is True:
         check_sequence_parallel(model)
 
     return model

internlm/utils/parallel.py

@@ -105,14 +105,11 @@ def set_model_params_layer_name(model):
 
 def check_sequence_parallel(model):
     """
-    check whether the norm module has IS_SEQUENC_PARALLEL attribute.
+    check whether the norm module has IS_SEQUENCE_PARALLEL attribute.
-    when the sequence_parallel is True, the norm module should have the IS_SEQUENC_PARALLEL attribute
+    when the sequence_parallel is True, the norm module should have the IS_SEQUENCE_PARALLEL attribute
     to illustrate the norm should conduct the all-reduce for its grad.
     """
 
-    if gpc.config.parallel.sequence_parallel is False:
-        return
-
     if not isinstance(model, nn.ModuleList):
         model = [model]
 
@@ -124,6 +121,6 @@ def check_sequence_parallel(model):
             if isinstance(module, (RMSNorm, nn.LayerNorm)):
                 for param in module.parameters():
                     assert hasattr(param, IS_SEQUENCE_PARALLEL), (
-                        "when the sequence parallel is True,"
+                        "when the gpc.config.parallel.sequence parallel is True,"
                         "the params of norm module should have IS_SEQUENCE_PARALLEL attribute"
                     )