Error when exporting Llama3.2-3B-Instruct for Android using QNN

[moxprox]

I have setup QNN backend and I am able to export sample model with it. However when exporting Llama-3.2-3B-Instruct for Android SA8295 with this command:

export KV_UPDATER="smart_mask"
export LLAMA_P="/home/moxprox/execu-torch-dev/models/Llama-3.2-3B-Instruct/original/"
python executorch/examples/qualcomm/oss_scripts/llama/llama.py \
-b build-android/ -m SA8295 --ptq 16a4w \
--checkpoint ${LLAMA_P}/consolidated.00.pth \
--params ${LLAMA_P}/params.json \
--tokenizer_model ${LLAMA_P}/tokenizer.model \
--prefill_ar_len 64 --max_seq_len 1024 --prompt "Once upon a time" \
--compile_only --llama_model llama3_2 --model_mode hybrid \
--kv_updater ${KV_UPDATER} --num_sharding 4

I get this weird error:

    flat_args_with_path = _check_inputs_match(args, kwargs, self._in_spec)
  File "/root/miniconda3/envs/executorch/lib/python3.10/site-packages/torch/export/_unlift.py", line 63, in _check_inputs_match
    raise ValueError(  # noqa: B904
torch._dynamo.exc.InternalTorchDynamoError: ValueError: Trying to flatten user inputs with exported input tree spec:
TreeSpec(tuple, None, [TreeSpec(tuple, None, [*,
    *,
    *,
    *]),
  TreeSpec(dict, [], [])])
but actually got inputs with tree spec of:
TreeSpec(tuple, None, [TreeSpec(tuple, None, [*,
    *,
    *,
    *,
    *,
    *]),
  TreeSpec(dict, [], [])]).
Please check that the inputs have the same number and type of args and kwargs as the ones you used when tracing.

from user code:
   File "/root/executorch/backends/qualcomm/utils/utils.py", line 1036, in forward
    ret = lower_module(*inp)
  File "/root/miniconda3/envs/executorch/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
    return forward_call(*args, **kwargs)
  File "/root/executorch/exir/lowered_backend_module.py", line 359, in forward
    return executorch_call_delegate(self, *args)

I cloned the model from HuggingFace, no changes have been made to model config. Could you provide some insight on how to fix this issue?
Thanks in advance

[moxprox]

• Im on:
• commit 360c9bb (HEAD -> viable/strict, origin/viable/strict)

[moxprox]

As suggested by @cccclai on discord, I moved to latest in main and the issue was fixed. Thanks.