[recipe] fix: remove redundant gradient_accumulation_fusion override in NT3 Nano

Now that the checker correctly returns True when transformer_engine.pytorch
is available, MambaModelProvider's default_factory handles this correctly.
The explicit override is no longer needed.

Signed-off-by: Yu Yao <yaoyu.094@gmail.com>
Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>

Author: yaoyu-33

src/megatron/bridge/recipes/nemotronh/nemotron_3_nano.py

@@ -187,7 +187,6 @@ def nemotron_3_nano_pretrain_config() -> ConfigContainer:
 
     cfg.model.init_method_std = 0.0173
     cfg.model.apply_rope_fusion = False
-    cfg.model.gradient_accumulation_fusion = True
     cfg.model.use_fused_weighted_squared_relu = True
 
     return cfg
@@ -300,7 +299,6 @@ def nemotron_3_nano_sft_config() -> ConfigContainer:
     cfg.model.moe_grouped_gemm = True
     cfg.model.cross_entropy_loss_fusion = True
     cfg.model.cross_entropy_fusion_impl = "native"
-    cfg.model.gradient_accumulation_fusion = True
 
     # Memory saving (recompute & offloading)
     cfg.model.recompute_granularity = None
@@ -490,7 +488,6 @@ def nemotron_3_nano_peft_config(
     cfg.model.moe_grouped_gemm = True
     cfg.model.cross_entropy_loss_fusion = True
     cfg.model.cross_entropy_fusion_impl = "native"
-    cfg.model.gradient_accumulation_fusion = True
 
     # Memory saving
     cfg.model.recompute_granularity = None