Qwen1_5/run_demo.sh

#!/bin/bash
set -ex

# download bmodel
if [ ! -d "../../bmodels" ]; then
 mkdir ../../bmodels
fi

if [ ! -f "../../bmodels/qwen1.5-1.8b_int4_1dev.bmodel" ]; then
 pip3 install dfss
 python3 -m dfss [email protected]:/ext_model_information/LLM/LLM-TPU/qwen1.5-1.8b_int4_1dev.bmodel
 mv qwen1.5-1.8b_int4_1dev.bmodel ../../bmodels
else
 echo "Bmodel Exists!"
fi

if [ ! -f "./python_demo/chat_parallel.cpython-38-aarch64-linux-gnu.so" ]; then
 cd python_demo && rm -rf build && mkdir build && cd build
 cmake .. && make -j
 cp chat_parallel.cpython-38-aarch64-linux-gnu.so ..
 cd ../..
else
 echo "chat.so exists!"
fi

# run demo
echo $PWD
python3 python_demo/pipeline.py --model ../../bmodels/qwen1.5-1.8b_int4_1dev.bmodel --tokenizer ./token_config --devid 0