[Hotfix] hotfix normalization (#6163)

* [fix] hotfix normalization

* [hotfix] force doc ci test

* [hotfix] fallback doc

colossalai/shardformer/layer/normalization.py

@@ -76,6 +76,7 @@ if SUPPORT_NPU:
 
     FusedRMSNormWithHook = NPUFusedRMSNormWithHook
 else:
+    try:
         from apex.normalization import FusedRMSNorm as ApexFusedRMSNorm
 
         class CUDAFusedRMSNormWithHook(ApexFusedRMSNorm):
@@ -88,6 +89,11 @@ else:
                 return output
 
         FusedRMSNormWithHook = CUDAFusedRMSNormWithHook
+    except ImportError:
+        warnings.warn(
+            "Please install apex from source (https://github.com/NVIDIA/apex) to use the fused RMSNorm kernel"
+        )
+
 
 FAST_LAYERNORM_SUPPORTED_SIZE = [
     1024,