[train]: fix bug of deepseek_v4 support pr. (#47)

# Description
Bug fix
PR Description
It is missed in previous deepseek_v4 support pr. And it will cause error
when moe forwarding. It is only used in moe bias_act_func currently.
## Type of change

- [ ] New feature (non-breaking change which adds functionality)
- [ ] Infra/Build change (changes to CI/CD workflows or build scripts)
- [ ] Code refactoring
- [ ] Documentation change
- [✅ ] Bug fix
- [ ] Breaking change

## Changes

- Content 1
- Content 2
- Content 3
- Content 4

## Checklist

- [ ] I have read and followed the contributing guidelines
- [ ] The functionality is complete
- [ ] I have commented my code, particularly in coverage report
uploading steps
- [ ] I have made corresponding changes to the documentation
- [ ] My changes generate no new warnings
- [ ] I have added/updated tests that prove my feature works
- [ ] New and existing unit tests pass locally

Author: LiJunscs

megatron/core/transformer/transformer_config.py

@@ -939,6 +939,15 @@ class TransformerConfig(ModelParallelConfig):
     """Number of SMs to use for HybridEP. In pure NVL scenarios,
     16 SMs can generally achieve good bandwidth."""
 
+    moe_mlp_glu_interleave_size: Optional[int] = None
+    """When set, GLU activations in the MoE grouped MLP layer will use a
+    block interleaved format. Instead of interpreting the input tensor
+    as a concatenation of gates and linear units, it will be
+    interpreted as alternating blocks of gates and linear units.
+
+    This data format is experimental and primarily intended to enable
+    advanced fused kernels."""
+
     ##################
     # Context Parallel
     ##################