Can't export a model (mms-tts-eng model from HuggingFace)

🐛 Describe the bug

The mms-tts-eng model from HuggingFace(https://huggingface.co/facebook/mms-tts-eng) doesn't export. The script reproducing the error is:

from transformers import VitsModel, AutoTokenizer
import torch

model = VitsModel.from_pretrained("facebook/mms-tts-eng")
model = model.eval()
tokenizer = AutoTokenizer.from_pretrained("facebook/mms-tts-eng")

text = "some example text in the English language"
inputs = tokenizer(text, return_tensors="pt")

print(inputs,inputs['input_ids'].shape,inputs['input_ids'].dtype,inputs['attention_mask'].shape,inputs['attention_mask'].dtype)

example_inputs = (inputs['input_ids'],inputs['attention_mask'])
exported_program = torch.export.export(model, example_inputs)

and the error is

item: "Sym(Eq(u2, 1))" = torch.ops.aten.item.default(ne);  ne = item = None
.....
torch.fx.experimental.symbolic_shapes.GuardOnDataDependentSymNode: Could not guard on data-dependent expression Eq(u2, 1) (unhinted: Eq(u2, 1)).  (Size-like symbols: none)

consider using data-dependent friendly APIs such as guard_or_false, guard_or_true and statically_known_true.
Caused by: (_export/non_strict_utils.py:1159 in __torch_function__)
...

I'm using ExecuTorch installed from source in editable mode and transformers 5.0.0.rc1.

It would be great if there was a way to export the model as it's an interesting use case, the NN is already trained in PyTorch and available in HuggingFace. I would like to deploy the model with ExecuTorch.
Is there a way to export the mms-tts-eng model ?

Versions

Environment information:

PyTorch version: 2.11.0.dev20260215
Is debug build: False
CUDA used to build PyTorch: None
ROCM used to build PyTorch: N/A

OS: macOS 15.7.4 (arm64)
GCC version: Could not collect
Clang version: 17.0.0 (clang-1700.6.4.2)
CMake version: version 3.31.6
Libc version: N/A

Python version: 3.10.16 (main, Dec  3 2024, 17:27:57) [Clang 16.0.0 (clang-1600.0.26.4)] (64-bit runtime)
Python platform: macOS-15.7.4-arm64-arm-64bit
Is CUDA available: False
CUDA runtime version: No CUDA
CUDA_MODULE_LOADING set to: N/A
GPU models and configuration: No CUDA
Nvidia driver version: No CUDA
cuDNN version: No CUDA
Is XPU available: False
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True
Caching allocator config: N/A

CPU:
Apple M3 Pro

Versions of relevant libraries:
[pip3] executorch==1.2.0a0+096f10c
[pip3] flake8==6.1.0
[pip3] flake8-breakpoint==1.1.0
[pip3] flake8-bugbear==24.4.26
[pip3] flake8-comprehensions==3.14.0
[pip3] flake8-plugin-utils==1.3.3
[pip3] flake8-pyi==23.5.0
[pip3] mypy==1.14.1
[pip3] mypy_extensions==1.1.0
[pip3] numpy==2.2.6
[pip3] onnxruntime==1.23.2
[pip3] optimum-executorch==0.2.0.dev0
[pip3] pytorch_tokenizers==1.1.0
[pip3] torch==2.11.0.dev20260215
[pip3] torchao==0.16.0+git026b76d12
[pip3] torchaudio==2.11.0.dev20260215
[pip3] torchcodec==0.10.0.dev20251222
[pip3] torchdata==0.11.0
[pip3] torcheval==0.0.7
[pip3] torchsr==1.0.4
[pip3] torchtune==0.0.0
[pip3] torchvision==0.26.0.dev20260215
[conda] Could not collect

cc @digantdesai @freddan80 @per @zingo @oscarandersson8218 @mansnils @Sebastian-Larsson @robell

Start by running the provided VitsModel and AutoTokenizer reproducer with the listed PyTorch, Transformers, and ExecuTorch versions, focusing on the torch.export.export entry point and the data-dependent Eq(u2, 1) failure. Done means the facebook/mms-tts-eng model exports successfully with the shown inputs and remains usable for ExecuTorch deployment.