Exporting .pte with mediatek backend fail

[matt0823]

Problem Description

I tried to run the LLaMA3.2 3B model on the MTK platform (D9400) and encountered an error when exporting the .pte model based on the instructions in examples/mediatek/README.md.

Command Used:

source shell_scripts/export_llama.sh llama3.2-3b 4 128 512 None

Error Log

Loading weights from disk
Shape: 128t512c
Shape: 1t512c
export shapes: {'128t512c': [128, 512], '1t512c': [1, 512]}
Max Num Token: 128
Max Cache Size: 512
Instantiating submodels
Loading weights for chunk 0
Loading weights for chunk 1
Loading weights for chunk 2
Loading weights for chunk 3
Exporting Chunk 0 to PTE
Getting pre autograd ATen Dialect Graph
Getting ATen Dialect Graph for Llama-3.2-3B-Instruct_A16W4_dummy_cal_4_chunks 128t512c chunk 0
Getting ATen Dialect Graph for Llama-3.2-3B-Instruct_A16W4_dummy_cal_4_chunks 1t512c chunk 0

Delegating Edge Program to Neuropilot Backend

Traceback (most recent call last):
  File "/mnt/f/Code/executorch/examples/mediatek/model_export_scripts/llama.py", line 499, in <module>
    main()
  File "/mnt/f/Code/executorch/examples/mediatek/model_export_scripts/llama.py", line 485, in main
    export_to_et_ir(
  File "/mnt/f/Code/executorch/examples/mediatek/model_export_scripts/llama.py", line 369, in export_to_et_ir
    delegated_program = to_backend(
  File "/usr/lib/python3.10/functools.py", line 889, in wrapper
    return dispatch(args[0].__class__)(*args, **kw)
  File "/home/matt823/.local/lib/python3.10/site-packages/executorch/exir/backend/backend_api.py", line 762, in _
    lower_all_submodules_to_backend(
  File "/home/matt823/.local/lib/python3.10/site-packages/executorch/exir/backend/backend_api.py", line 591, in lower_all_submodules_to_backend
    backend_name_to_subclass[backend_id].preprocess_multimethod(
  File "/home/matt/.local/lib/python3.10/site-packages/executorch/backends/mediatek/preprocess.py", line 195, in preprocess_multimethod
    blob, new_models = mtk_neuron.extract_shared_data(
AttributeError: module 'mtk_neuron' has no attribute 'extract_shared_data'

Additional Information

I checked the mtk_neuron module and confirmed it does not have an attribute named extract_shared_data. A similar interface named extract_shared_hint does exist.

The tools were downloaded from:
https://s3.ap-southeast-1.amazonaws.com/mediatek.neuropilot.com/06302508-4c94-4bf2-9789-b0ee44e83e27.gz

Request for Assistance
Could you please help investigate this issue? It appears there might be a mismatch between the expected API and the actual implementation in the mtk_neuron module. Thank you.