Add qwen3 meta loading for AutoTP (deepspeedai#7293)

This PR fixes deepspeedai#7275 to
enable Qwen3 meta loading for AutoTP

Signed-off-by: Ma, Guokai <guokai.ma@intel.com>
Signed-off-by: Max Kovalenko <mkovalenko@habana.ai>

Author: delock

deepspeed/module_inject/auto_tp.py

@@ -137,8 +137,8 @@ def is_load_module(module):
             "LPLayerNorm", "SharedEmbedding", "OPTLearnedPositionalEmbedding", "LlamaRMSNorm", "FalconLinear",
             "MistralRMSNorm", "T5LayerNorm", "MixtralRMSNorm", "Phi3RotaryEmbedding", "Phi3SuScaledRotaryEmbedding",
             "Phi3RMSNorm", "YuanRMSNorm", "YuanRotaryEmbedding", "Phi3LongRoPEScaledRotaryEmbedding", "Qwen2RMSNorm",
-            "DeepseekV2RMSNorm", "DeepseekV3RMSNorm", "DeepseekV2YarnRotaryEmbedding", "DeepseekV3YarnRotaryEmbedding",
-            "MoEGate"
+            "Qwen3RMSNorm", "DeepseekV2RMSNorm", "DeepseekV3RMSNorm", "DeepseekV2YarnRotaryEmbedding",
+            "DeepseekV3YarnRotaryEmbedding", "MoEGate"
         ]
         return module.__class__ in load_layers or module._get_name() in load_layer_names