Follow the guide of qnn Android demo to run until coredump occurs during qnn model quantization

[AndreaChiChengdu]

🐛 Describe the bug

https://github.com/pytorch/executorch/blob/main/examples/demo-apps/android/LlamaDemo/docs/delegates/qualcomm_README.md
I followed this demo to start quantifying the model. The beginning was smooth, but the flatbuffer assertion would occur at the end. My model was llama2-7b and quantized by a16w4, which could not generate the final pte for qnn. I've also tried --num_sharding 4, but other problems pop up

PTQ# 4 bits weight only quantize
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_16a4w -d fp32 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_name="test.pte”
It starts out fine, but ends up with the following problem
————————————————————————————————————————————————————

INFO:executorch.backends.qualcomm.qnn_preprocess:Visiting: aten_convolution_default_224, aten.convolution.default
INFO:executorch.backends.qualcomm.qnn_preprocess:Visiting: aten_permute_copy_default_3493, aten.permute_copy.default
INFO:executorch.backends.qualcomm.qnn_preprocess:Visiting: aten_view_copy_default_577, aten.view_copy.default
INFO:executorch.backends.qualcomm.qnn_preprocess:Visiting: quantized_decomposed_dequantize_per_tensor_tensor, quantized_decomposed.dequantize_per_tensor.tensor
python: /002data/andrea/executorch/qnn/executorch/third-party/flatbuffers/include/flatbuffers/vector_downward.h:146: size_t flatbuffers::vector_downward::ensure_space(size_t) [with SizeT = unsigned int; size_t = long unsigned int]: Assertion `size() < max_size_' failed.
Aborted (core dumped)

Versions

source code build v0.5 follow the android llama demo for qualcomm