Skip to content

Some doubts regarding quantization in academia and industry. #32

@BambooKui

Description

@BambooKui

I have some questions. Previously, I quantized some models using torch.fx, and both the size and speed of the models changed. However, I noticed that these tools are not used in academia. So, I used your code to learn how to quantize a model in an academic context, but it seems that there is no change in the size and speed of the model, and the stored data remains unchanged as well. Is this normal? And is it not necessary to fuse operators or similar elements?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions