Torch-to-Linalg lowering fails to legalize torch.vtensor.literal for Hugging Face TinyLlama forward export

Summary

I am trying to lower a Hugging Face transformer-style model through Torch-MLIR to LINALG_ON_TENSORS, but the lowering fails during the Torch backend IR → Linalg-on-tensors conversion with:

failed to legalize operation 'torch.vtensor.literal'

usr/lib/python3.12/copyreg.py:99: FutureWarning: `isinstance(treespec, LeafSpec)` is deprecated, use `isinstance(treespec, TreeSpec) and treespec.is_leaf()` instead.
  return cls.__new__(cls, *args)
loc(callsite("/home/ubuntu/sa_mlir_full_stack/venv/lib/python3.12/site-packages/transformers/models/llama/modeling_llama.py":399:0 at callsite("/home/ubuntu/sa_mlir_full_stack/venv/lib/python3.12/site-packages/transformers/models/llama/modeling_llama.py":474:0 at "/home/ubuntu/models_mod_2/capture_frontend.py":80:0))): error: failed to legalize operation 'torch.vtensor.literal'
Traceback (most recent call last):
  File "/home/ubuntu/models_mod_2/compile_to_hazard3.py", line 628, in <module>
    main()
  File "/home/ubuntu/models_mod_2/compile_to_hazard3.py", line 468, in main
    mlir_module = export_and_import(
                  ^^^^^^^^^^^^^^^^^^
  File "/home/ubuntu/sa_mlir_full_stack/venv/lib/python3.12/site-packages/torch_mlir/fx.py", line 147, in export_and_import
    return _module_lowering(
           ^^^^^^^^^^^^^^^^^
  File "/home/ubuntu/sa_mlir_full_stack/venv/lib/python3.12/site-packages/torch_mlir/fx.py", line 86, in _module_lowering
    return lower_mlir_module(verbose, output_type, torch_mod, backend_options)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/ubuntu/sa_mlir_full_stack/venv/lib/python3.12/site-packages/torch_mlir/compiler_utils.py", line 218, in lower_mlir_module
    run_pipeline_with_repro_report(
  File "/home/ubuntu/sa_mlir_full_stack/venv/lib/python3.12/site-packages/torch_mlir/compiler_utils.py", line 136, in run_pipeline_with_repro_report
    raise TorchMlirCompilerError(trimmed_message) from None
torch_mlir.compiler_utils.TorchMlirCompilerError: Lowering Torch Backend IR -> Linalg-on-Tensors Backend IR failed with the following diagnostics:
 
 
python exception: Failure while executing pass pipeline
 
For Torch-MLIR developers, the error can be reproduced with:
$ torch-mlir-opt -pass-pipeline='builtin.module(torch-backend-to-linalg-on-tensors-backend-pipeline{allow-non-finites=True})' /tmp/UnnammedModule.mlir
Add '-mlir-print-ir-after-all -mlir-disable-threading' to get the IR dump for debugging purpose.

Please let me know if you need more context. @momchil-velikov 

Reproduce the failure with the torch-backend-to-linalg-on-tensors-backend-pipeline command and inspect /tmp/UnnammedModule.mlir. Use -mlir-print-ir-after-all and -mlir-disable-threading as suggested, starting from the torch.vtensor.literal operation and the call sites in capture_frontend.py and modeling_llama.py. Done means the TinyLlama forward export passes this lowering pipeline without the legalization error.