🐛 [Bug] dynamic_block_quantize converter received an input of (11520, 3, 2, 16, 16) shape. Supported shapes: 2D or 3D

[zewenli98]

Bug Description

I got the error when quantizing Alpamayo 1 with nvfp4:

2026-04-21 18:01:13,791 - torch_tensorrt.dynamo.conversion._TRTInterpreter - INFO - Converted node hidden_states [hidden_states] (Inputs: () | Outputs: (hidden_states: (11520, 1536)@torch.float16))
2026-04-21 18:01:13,791 - torch_tensorrt.dynamo.conversion._TRTInterpreter - INFO - Converted node visual.patch_embed/_reshape_copy [aten._reshape_copy.default] (Inputs: (hidden_states: (11520, 1536)@torch.float16, [-1, 3, 2, 16, 16]) | Outputs: (_reshape_copy: (11520, 3, 2, 16, 16)@torch.float16))
2026-04-21 18:01:13,792 - torch_tensorrt.dynamo.conversion._TRTInterpreter - INFO - Converted node visual_patch_embed_proj_input_quantizer__amax [visual.patch_embed.proj.input_quantizer._amax] (Inputs: () | Outputs: (visual_patch_embed_proj_input_quantizer__amax: ()@torch.float16))
2026-04-21 18:01:13,795 - alpamayo_r1.trt.vision - ERROR - TRT compilation failed: dynamic_block_quantize converter received an input of (11520, 3, 2, 16, 16) shape. Supported shapes: 2D or 3D

While executing %dynamic_block_quantize_op : [num_users=1] = call_function[target=torch.ops.tensorrt.dynamic_block_quantize_op.default](args = (%_reshape_copy, 16, %visual_patch_embed_proj_input_quantizer__amax, 4, 2, 8, 4), kwargs = {})
Original traceback:
File "/home/scratch.zewenl_sw/docker_workspace/cehongwang/alpamayo/src/alpamayo_r1/trt/vision.py", line 192, in forward
    hidden_states = self.visual.patch_embed(hidden_states)
  File "/home/scratch.zewenl_sw/docker_workspace/alpamayo/ar1_venv-b300/lib/python3.12/site-packages/transformers/models/qwen3_vl/modeling_qwen3_vl.py", line 75, in forward
    hidden_states = self.proj(hidden_states.to(dtype=target_dtype)).view(-1, self.embed_dim)
  File "/home/scratch.zewenl_sw/docker_workspace/tmp/TensorRT-Model-Optimizer/modelopt/torch/quantization/nn/modules/quant_module.py", line 232, in forward
    return super().forward(input, *args, **kwargs)
  File "/home/scratch.zewenl_sw/docker_workspace/tmp/TensorRT-Model-Optimizer/modelopt/torch/quantization/nn/modules/quant_module.py", line 164, in forward
    input = self.input_quantizer(input)
  File "/home/scratch.zewenl_sw/docker_workspace/tmp/TensorRT-Model-Optimizer/modelopt/torch/quantization/nn/modules/tensor_quantizer.py", line 1086, in forward
    outputs = self._fake_quantize(inputs)
  File "/home/scratch.zewenl_sw/docker_workspace/tmp/TensorRT-Model-Optimizer/modelopt/torch/quantization/tensor_quant.py", line 552, in forward
    return _dynamic_block_quantize_forward(
Use tlparse to see full graph. (https://github.com/pytorch/tlparse?tab=readme-ov-file#tlparse-parse-structured-pt2-logs)
2026-04-21 18:01:13,796 - alpamayo_r1.trt.compile_trt - ERROR - Vision TRT compilation failed

To Reproduce

Expected behavior

Environment

Build information about Torch-TensorRT can be found by turning on debug messages

• Torch-TensorRT Version (e.g. 1.0.0):
• PyTorch Version (e.g. 1.0):
• CPU Architecture:
• OS (e.g., Linux):
• How you installed PyTorch (conda, pip, libtorch, source):
• Build command you used (if compiling from source):
• Are you using local sources or building from archives:
• Python version:
• CUDA version:
• GPU models and configuration:
• Any other relevant information:

Additional context