fix: Add get_hf_tokenizer_kwargs to NemotronHBridge

src/megatron/bridge/models/nemotronh/nemotron_h_bridge.py

@@ -104,6 +104,15 @@ def provider_bridge(self, hf_pretrained: PreTrainedCausalLM) -> MambaModelProvid
 
         return provider
 
+    @classmethod
+    def get_hf_tokenizer_kwargs(cls) -> dict:
+        """Return HuggingFace tokenizer kwargs for Nemotron-H models.
+
+        Nemotron-H models only provide a fast tokenizer (tokenizer.json),
+        so use_fast=True is required.
+        """
+        return {"use_fast": True}
+
     def mapping_registry(self) -> MegatronMappingRegistry:
         # Return MegatronMappingRegistry containing parameter mappings from Megatron to HF format
         # First create simple 1:1 parameter mappings using a dictionary for readability

tests/unit_tests/models/nemotronh/test_nemotron_h_bridge.py

@@ -357,6 +357,20 @@ def test_mapping_registry_moe_hf_params(self):
         )
 
 
+class TestNemotronHBridgeTokenizerKwargs:
+    """Test get_hf_tokenizer_kwargs method."""
+
+    def test_tokenizer_kwargs_returns_dict(self):
+        """Test get_hf_tokenizer_kwargs returns a dict."""
+        kwargs = NemotronHBridge.get_hf_tokenizer_kwargs()
+        assert isinstance(kwargs, dict)
+
+    def test_tokenizer_kwargs_use_fast(self):
+        """Test get_hf_tokenizer_kwargs returns use_fast=True."""
+        kwargs = NemotronHBridge.get_hf_tokenizer_kwargs()
+        assert kwargs.get("use_fast") is True
+
+
 class TestAutoBridgeIntegration:
     """Integration tests for AutoBridge with NemotronH models."""