[BUG]Megatron-LM doesn't support transformer-engine 1.13

[klhhhhh]

Describe the bug
Megatron-LM doesn't compatible with transformer-engine 1.13.

in transformer-engine:

https://github.com/NVIDIA/TransformerEngine/blob/2643ba1df43397cc84c9da5fe719a66d87ad9a0a/transformer_engine/pytorch/module/layernorm.py#L62-L70

new parameter normalized_shape was declared, but in Megatron-LM:

Megatron-LM/megatron/core/extensions/transformer_engine.py

Lines 64 to 70 in 1b8fce7

	instance = te.pytorch.LayerNorm(
	hidden_size=hidden_size,
	eps=eps,
	sequence_parallel=config.sequence_parallel,
	zero_centered_gamma=config.layernorm_zero_centered_gamma,
	**_get_extra_te_kwargs(config),
	)

the new parameter normalized_shape is missing.

To Reproduce
When directly installing transformer-engine with

pip install git+https://github.com/NVIDIA/TransformerEngine.git@stable

transformer engine version will be 1.13, and when you run the scripts in examples, the error will appear.

Environment (please complete the following information):

• Megatron-LM commit ID: 1b8fce7e17e7f945c1f59d06744a2e126bedc015
• PyTorch version: 2.5
• CUDA version: 12.4
• NCCL version: 2.21.5
• transformer-engine version: 1.13

Fix right now:
install transformer-engine with:

pip install git+https://github.com/NVIDIA/TransformerEngine.git@release_v1.12

[timmoon10]

Thanks for reporting. This regression was introduced in NVIDIA/TransformerEngine#1033 and it should be fixed with NVIDIA/TransformerEngine#1329. Hopefully the integration test in that PR will help catch these kinds of errors in the future.