diff --git a/src/megatron/bridge/models/qwen/qwen_provider.py b/src/megatron/bridge/models/qwen/qwen_provider.py
index f211c0c84c..2a7b6fd735 100644
--- a/src/megatron/bridge/models/qwen/qwen_provider.py
+++ b/src/megatron/bridge/models/qwen/qwen_provider.py
@@ -19,7 +19,9 @@
 
 import torch
 import torch.nn.functional as F
-from megatron.core.models.gpt.gpt_layer_specs import get_gpt_decoder_block_spec
+from megatron.core.models.gpt.experimental_attention_variant_module_specs import (
+    get_transformer_block_with_experimental_attention_variant_spec,
+)
 from megatron.core.transformer.spec_utils import ModuleSpec
 
 from megatron.bridge.models.gpt_provider import GPTModelProvider
@@ -433,8 +435,8 @@ class Qwen3MoEModelProvider235B_A22B(Qwen3MoEModelProvider):
 class Qwen3NextModelProvider(Qwen3MoEModelProvider):
     """Base provider for Qwen 3 Next Models."""
 
-    transformer_layer_spec: ModuleSpec | Callable[["GPTModelProvider"], ModuleSpec] = partial(
-        get_gpt_decoder_block_spec, use_transformer_engine=HAVE_TE
+    transformer_layer_spec: ModuleSpec | Callable[["GPTModelProvider"], ModuleSpec] = (
+        get_transformer_block_with_experimental_attention_variant_spec
     )
 
     layernorm_zero_centered_gamma: bool = True  # Zero-centered RMSNorm
diff --git a/src/megatron/bridge/training/mlm_compat/model.py b/src/megatron/bridge/training/mlm_compat/model.py
index 7bd088c44c..e32153956d 100644
--- a/src/megatron/bridge/training/mlm_compat/model.py
+++ b/src/megatron/bridge/training/mlm_compat/model.py
@@ -21,6 +21,9 @@
 from megatron.core.enums import ModelType
 from megatron.core.fp8_utils import correct_amax_history_if_needed
 from megatron.core.models.gpt import GPTModel
+from megatron.core.models.gpt.experimental_attention_variant_module_specs import (
+    get_transformer_block_with_experimental_attention_variant_spec,
+)
 from megatron.core.models.gpt.gpt_layer_specs import (
     get_gpt_decoder_block_spec,
     get_gpt_layer_local_spec,
@@ -90,7 +93,13 @@ def _gpt_provider(
     if config is None:
         config = _transformer_config_from_args(args)
 
-    if args.num_experts:
+    if args.experimental_attention_variant is not None:
+        transformer_layer_spec = (
+            get_transformer_block_with_experimental_attention_variant_spec(
+                config=config, vp_stage=vp_stage
+            )
+        )
+    elif args.num_experts:
         # Define the decoder block spec
         transformer_layer_spec = get_gpt_decoder_block_spec(
             config,