wasi_nn_udf

How to use:

Install WasmEdge and then install the WasiNN plugin as follows.

# Install PyTorch
PYTORCH_VERSION="1.8.2"
curl -s -L -O --remote-name-all "https://download.pytorch.org/libtorch/lts/1.8/cpu/libtorch-cxx11-abi-shared-with-deps-${PYTORCH_VERSION}%2Bcpu.zip"
unzip -q "libtorch-cxx11-abi-shared-with-deps-${PYTORCH_VERSION}%2Bcpu.zip"
rm -f "libtorch-cxx11-abi-shared-with-deps-${PYTORCH_VERSION}%2Bcpu.zip"
export LD_LIBRARY_PATH=$(pwd)/libtorch/lib:${LD_LIBRARY_PATH}
export Torch_DIR=$(pwd)/libtorch

# Download and extract the plugin
wget "https://github.com/WasmEdge/WasmEdge/releases/download/0.12.0-alpha.2/WasmEdge-plugin-wasi_nn-pytorch-0.12.0-alpha.2-ubuntu20.04_x86_64.tar.gz"

tar -xzf "WasmEdge-plugin-wasi_nn-pytorch-0.12.0-alpha.2-ubuntu20.04_x86_64.tar.gz"

# Install the plugin if your wasmedge is installed in ~/.wasmedge
cp libwasmedgePluginWasiNN.so ~/.wasmedge/plugin/

# Install the plugin if your wasmedge is installed in /usr/local
cp libwasmedgePluginWasiNN.so /usr/local/lib/wasmedge/

Make sure that you build libsql with WasmEdge support.

git clone https://github.com/libsql/libsql
cd libsql
./configure --enable-wasm-runtime-wasmedge
make

Build the WASI NN UDF example.

cargo wasi build --release

Create test.sql file for libsql and run libsql

./gen_demo_sql.sh

libsql

Execute in libsql

> .read test.sql

Note

The gen_demo_sql.sh script

• first converts the image file to a tensor file. This is not strictly necessary as the UDF itself can perform this conversion.
• then calls gen_insert_image_sql.sh to create a SQL file that inserts the tensor file into a database table as a blob.
• then creates a database table with the above mentioned blob field, and calls the generated SQL file to insert the blob.
• then calls gen_libsql_udf.sh to create the UDF. The PyTorch model is embedded in the UDF.
• finally, uses the UDF to classify the blob in a SQL query.