Does Quantized ONNX Model Always Start with QuantizeLinear as the First Layer?

[nirdesh17]

I have a question about the structure of quantized ONNX models.

When we quantize an ONNX model to int8 or uint8 using static quantization, is it guaranteed that the first layer of the quantized model will always be a QuantizeLinear node? Or does this depend on the specific quantization method or the tool used for the quantization process?

I’m trying to understand whether this is a standardized behavior for models quantized with static quantization or if it varies based on implementation details.

Any insights, explanations, or references to relevant documentation would be greatly appreciated.

Thank you!