[Python] UserWarning: Mutation on a buffer in the model is detected. ExecuTorch assumes...

I have exported model to Executorch program, and the file is saved correctly in directory. But I enountered warning like this

# To see more debug info, please use `graph_module.print_readable()`
Model successfully exported to model.pte
/usr/local/lib/python3.10/dist-packages/executorch/exir/emit/_emitter.py:1512: UserWarning: Mutation on a buffer in the model is detected. ExecuTorch assumes buffers that are mutated in the graph have a meaningless initial state, only the shape and dtype will be serialized.
warnings.warn(


Yes, the warning was not completed, so I was confused about whether this warning would have a big impact on the program I was going to run or not.

I use T4 in google colab to execute this function.


def quantize_model(model: nn.Module, example_inputs: tuple) -> nn.Module:

print(f"Original model architecture:\n{model}")

# Initialize quantizer with symmetric quantization configuration
quantizer = XNNPACKQuantizer()
operator_config = get_symmetric_quantization_config(is_per_channel=False)
quantizer.set_global(operator_config)

# Prepare model for quantization
pre_autograd_model = export_for_training(model, example_inputs).module()
prepared_model = prepare_pt2e(pre_autograd_model, quantizer)

# Calibrate and convert the model
prepared_model(*example_inputs)
quantized_model = convert_pt2e(prepared_model)

print(f"Quantized model architecture:\n{quantized_model}")
return quantized_model

def export_to_executorch(model: nn.Module, example_inputs: tuple, output_path: str):

edge_program = to_edge_transform_and_lower(
export(model, example_inputs),
compile_config=EdgeCompileConfig(_check_ir_validity=True),
partitioner=[XnnpackPartitioner()]
)

# Convert to executorch program
executorch_program: exir.ExecutorchProgramManager = edge_program.to_executorch(
ExecutorchBackendConfig(passes=[])
)

# Save the program to file
with open(output_path, "wb") as file:
file.write(executorch_program.buffer)
print(f"Model successfully exported to {output_path}")

def main():
# Define example inputs
example_inputs = (torch.randn(1, 1, 256, 256),)

# Create and quantize model
model = ResEmoteNet() # Assuming this class is defined elsewhere
quantized_model = quantize_model(model, example_inputs)

# Export the quantized model
export_to_executorch(
model=quantized_model,
example_inputs=example_inputs,
output_path="model.pte"
)

if __name__ == "__main__":
main()

Continue reading...