Skip to content

Use MTP as quantization benchmark #8

@JohannesGaessler

Description

@JohannesGaessler

One would intuitively expect that the next token predictions from MTP layers become worse as you quantize them. So by determining the acceptance rate vs. the size one can create a benchmark for quantization formats. This would not be directly comparable between models but it would be very cheap to run so it may be useful as a proxy for more expensive but better benchmarks.

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