Add NemotronH_Omni_Reasoning_V3 as a supported Nemotron architecture (#52929)

Signed-off-by: Nave Assaf <nassaf@nvidia.com>

tests/models/registry.py

@@ -1250,6 +1250,10 @@ _MULTIMODAL_EXAMPLE_MODELS = {
     "NemotronH_Super_Omni_Reasoning_V3": _HfExamplesInfo(
         "nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16", is_available_online=False
     ),
+    # TODO: Change repo id once pertinent archs are public.
+    "NemotronH_Omni_Reasoning_V3": _HfExamplesInfo(
+        "nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16", is_available_online=False
+    ),
     "OpenCUAForConditionalGeneration": _HfExamplesInfo(
         "xlangai/OpenCUA-7B",
         trust_remote_code=True,

vllm/config/speculative.py

@@ -490,7 +490,10 @@ class SpeculativeConfig:
                 {"n_predict": n_predict, "architectures": ["ErnieMTPModel"]}
             )
 
-        if hf_config.architectures[0] == "NemotronH_Super_Omni_Reasoning_V3":
+        if hf_config.architectures[0] in (
+            "NemotronH_Super_Omni_Reasoning_V3",
+            "NemotronH_Omni_Reasoning_V3",
+        ):
             # Promote VLM's text_config so MTP detection below fires correctly
             hf_config = hf_config.text_config
 

vllm/model_executor/models/registry.py

@@ -535,6 +535,7 @@ _MULTIMODAL_MODELS = {
     "NemotronH_Nano_VL_V2": ("nano_nemotron_vl", "NemotronH_Nano_VL_V2"),
     "NemotronH_Nano_Omni_Reasoning_V3": ("nano_nemotron_vl", "NemotronH_Nano_VL_V2"),
     "NemotronH_Super_Omni_Reasoning_V3": ("nano_nemotron_vl", "NemotronH_Nano_VL_V2"),
+    "NemotronH_Omni_Reasoning_V3": ("nano_nemotron_vl", "NemotronH_Nano_VL_V2"),
     "NVLM_D": ("nvlm_d", "NVLM_D_Model"),
     "MuseGlimmerForConditionalGeneration": ("muse_glimmer", "MuseGlimmerForCausalLM"),
     "OpenCUAForConditionalGeneration": ("opencua", "OpenCUAForConditionalGeneration"),