PyTorch 모델을 스마트폰·마이크로컨트롤러 등 엣지 기기에 직접 배포하는 온디바이스 AI 런타임

pip install executorch

# Python 코드로 모델 export
python3 << 'EOF'
import torch
from executorch.exir import to_edge_transform_and_lower
from executorch.backends.xnnpack.partition.xnnpack_partitioner import XnnpackPartitioner

model = torch.nn.Linear(10, 5).eval()
example_inputs = (torch.randn(1, 10),)
exported_program = torch.export.export(model, example_inputs)
program = to_edge_transform_and_lower(
    exported_program,
    partitioner=[XnnpackPartitioner()]
).to_executorch()

with open("model.pte", "wb") as f:
    f.write(program.buffer)
print("Model exported to model.pte")
EOF

python3 << 'EOF'
from executorch.runtime import Runtime
import torch

runtime = Runtime.get()
method = runtime.load_program("model.pte").load_method("forward")
outputs = method.execute([torch.randn(1, 10)])
print("Output shape:", outputs[0].shape)
EOF

python -m executorch.extension.llm.export.export_llm --model llama3_2 --output llama.pte