I pulled the latest code, and the model is reporting errors everywhere

[yangh0597]

🐛 Describe the bug

I pulled the latest code, and the model is reporting errors everywhere. Two days ago it was fine.Looks like the code forgot to commit

command is:

 python -m examples.models.llama.export_llama --checkpoint "${MODEL_DIR}/consolidated.00.pth" -p "${MODEL_DIR}/params.json" -kv --disable_dynamic_shape --qnn --pt2e_quantize qnn_16a16w -d fp32 --num_sharding 4 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_name="Llama-Guard-3.2-1B-qnn_16a16w_s4.pte"

error is

Traceback (most recent call last):
  File "/opt/anaconda3/envs/et_qnn/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/opt/anaconda3/envs/et_qnn/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/opt/executorch/examples/models/llama/export_llama.py", line 32, in <module>
    main()  # pragma: no cover
  File "/opt/executorch/examples/models/llama/export_llama.py", line 28, in main
    export_llama(args)
  File "/opt/executorch/examples/models/llama/export_llama_lib.py", line 508, in export_llama
    builder = _export_llama(args)
  File "/opt/executorch/examples/models/llama/export_llama_lib.py", line 783, in _export_llama
    from executorch.backends.qualcomm.utils.utils import canonicalize_program
ImportError: cannot import name 'canonicalize_program' from 'executorch.backends.qualcomm.utils.utils' (/opt/executorch/backends/qualcomm/utils/utils.py)

I checked the/opt/executorch/backends/qualcomm/utils/utils. Py files, do not canonicalize_program this method

Versions

main

[metascroy]

Tagged this with QNN.

@yangh0597 I noticed that the executorch modules are not calling something from an anaconda environment (/opt/executorch/examples/models/llama/export_llama.py), werhereas runpy is (/opt/anaconda3/envs/et_qnn/lib/python3.10/runpy.py). Did you install ET to your anaconda environment after pulling in the latest code?

[yangh0597]

Tagged this with QNN.

@yangh0597 I noticed that the executorch modules are not calling something from an anaconda environment (/opt/executorch/examples/models/llama/export_llama.py), werhereas runpy is (/opt/anaconda3/envs/et_qnn/lib/python3.10/runpy.py). Did you install ET to your anaconda environment after pulling in the latest code? What is ET? I follow this document：https://github.com/pytorch/executorch/blob/main/examples/demo-apps/android/LlamaDemo/docs/delegates/qualcomm_README.md

[metascroy]

Yeah, it looks like canonicalize_program was removed from utils.py a couple days ago here: https://github.com/pytorch/executorch/commit/4086509b520dfbee3db86a033c262cb9248fedbd#diff-0439f6a7c1a3a3cfb222cd6409b6754f17a1ce782dd231de1d12bbf957d588f7L205

But this is imported in llama export here: https://github.com/pytorch/executorch/blob/main/examples/models/llama/export_llama_lib.py?lines=765

@haowhsu-quic, it looks like your PR https://github.com/pytorch/executorch/pull/6657 broke llama export for QNN, can you have a look?

cc @cccclai

[haowhsu-quic]

Hi @metascroy, sorry for the inconvenience. canonicalize_program was changed to https://github.com/pytorch/executorch/blob/b132c96ecf3cfc9d96358d0b8fe94de9426193f2/backends/qualcomm/utils/utils.py#L215 I wasn't aware of this part when submitting PR, will fire another one to fix it.

[scsonic]

Me too temp fix for me in utils.py

def canonicalize_program(obj):
    update_spill_fill_size(obj)