[bug] Qwen1.5-1.8B-Chat-MNN 模型更新后运行报错

[HRXWEB]

硬件

• M4 Pro

编译

mkdir build
cd build
cmake -DMNN_METAL=ON -DCMAKE_INSTALL_PREFIX=./install ..
make -j4

模型

Qwen1.5-1.8B-chat-MNN

在 2024.12.31 更新之前的模型版本可以用，但是 2024.12.31 更新后的模型运行报错：

运行

./cli_demo ~/.cache/modelscope/hub/MNN/Qwen1___5-1___8B-Chat-MNN/config.json
model path is /Users/xxx/.cache/modelscope/hub/MNN/Qwen1___5-1___8B-Chat-MNN/config.json
The device supports: i8sdot:1, fp16:1, i8mm: 1, sve2: 0
Can't open file:.tempcache
Load Cache file error.
### is_single_ = 1
load tokenizer
tokenizer_type = 3
load tokenizer Done
load /Users/xxx/.cache/modelscope/hub/MNN/Qwen1___5-1___8B-Chat-MNN/llm.mnn ... Load Module Done!
Clone Decode Module Done!

Q: nihao

A: [1]    18139 segmentation fault  ./cli_demo ~/.cache/modelscope/hub/MNN/Qwen1___5-1___8B-Chat-MNN/config.json

[HRXWEB]

ubuntu 20.04 也复现问题