[TRTLLM-6577][feat] Support nano_v2_vlm in pytorch backend

.github/CODEOWNERS

@@ -100,6 +100,8 @@
 /tests/unittest/_torch/modeling/test_modeling_pixtral.py @NVIDIA/trt-llm-torch-models-vlm-devs @NVIDIA/trt-llm-torch-models-devs
 
 ### TensorRT-LLM Pytorch - Models - Nemotron
+/tensorrt_llm/_torch/models/modeling_nanov2vlm.py @NVIDIA/trt-llm-torch-models-nemotron-devs @NVIDIA/trt-llm-torch-models-vlm-devs @NVIDIA/trt-llm-torch-models-devs
+/tensorrt_llm/_torch/models/modeling_radio.py @NVIDIA/trt-llm-torch-models-nemotron-devs @NVIDIA/trt-llm-torch-models-vlm-devs @NVIDIA/trt-llm-torch-models-devs
 /tensorrt_llm/_torch/models/modeling_nemotron_nas.py @NVIDIA/trt-llm-torch-models-nemotron-devs @NVIDIA/trt-llm-torch-models-devs
 /tensorrt_llm/_torch/models/modeling_nemotron_h.py @NVIDIA/trt-llm-torch-models-nemotron-devs @NVIDIA/trt-llm-torch-models-devs
 /tensorrt_llm/_torch/models/modeling_nemotron_nas.py @NVIDIA/trt-llm-torch-models-nemotron-devs @NVIDIA/trt-llm-torch-models-devs

cpp/kernels/fmha_v2/setup.py

@@ -1982,8 +1982,8 @@ def selected_mask_types(kspec):
             custom_mask = '0'
         # encoder models (head_size = 32 / 64 / 128) need packed_qkv input layout + padding mask.
         elif kspec.input_layout == InputLayout.PACKED_QKV:
-            # NOTE: 72 is added for vision transformer
-            if kspec.head_size not in [32, 64, 72, 128]:
+            # NOTE: 72/80 are added for vision transformer
+            if kspec.head_size not in [32, 64, 72, 80, 128]:
                 padding_mask = '0'
         # only cross attention (head_size = 32/64/128) needs contiguous_q_kv input layout + padding mask / custom_mask.
         elif kspec.input_layout == InputLayout.CONTIGUOUS_Q_KV:

docs/source/models/supported-models.md

@@ -51,6 +51,7 @@ Note: Support for other models may vary. Features marked "N/A" are not applicabl
 | LlavaNextForConditionalGeneration  | Yes               | Yes        | No              | Yes           | Yes              | No             | Yes                   | No                        | L + I    |
 | Llama4ForConditionalGeneration     | Yes               | Yes        | No              | Yes           | Yes              | No             | Yes                   | No                        | L + I    |
 | Mistral3ForConditionalGeneration   | Yes               | Yes        | Yes             | Yes           | Yes              | Yes            | Yes                   | No                        | L + I    |
+| NemotronH_Nano_VL_V2               | Yes               | Yes        | Yes             | Yes           | Yes              | No             | Yes                   | No                        | L + I + V |
 | Phi4MMForCausalLM                  | Yes               | Yes        | No              | Yes           | Yes              | No             | Yes                   | No                        | L + I + A |
 | Qwen2VLForConditionalGeneration    | Yes               | Yes        | No              | Yes           | Yes              | Yes            | Yes                   | No                        | L + I + V |
 | Qwen2_5_VLForConditionalGeneration | Yes               | Yes        | No              | Yes           | Yes              | Yes            | Yes                   | No                        | L + I + V |

tensorrt_llm/_torch/models/__init__.py

@@ -14,6 +14,7 @@
 from .modeling_llava_next import LlavaNextModel
 from .modeling_mistral import Mistral3VLM, MistralForCausalLM
 from .modeling_mixtral import MixtralForCausalLM
+from .modeling_nanov2vlm import NemotronH_Nano_VL_V2
 from .modeling_nemotron import NemotronForCausalLM
 from .modeling_nemotron_h import NemotronHForCausalLM
 from .modeling_nemotron_nas import NemotronNASForCausalLM
@@ -45,6 +46,7 @@
     "Mistral3VLM",
     "MistralForCausalLM",
     "MixtralForCausalLM",
+    "NemotronH_Nano_VL_V2",
     "NemotronForCausalLM",
     "NemotronHForCausalLM",
     "NemotronNASForCausalLM",