Skip to content

Compatability With GGUF format #41

@kimjoohyungsd

Description

@kimjoohyungsd

I've tried to apply SpinQuant to llama.cpp by following steps

  1. Fuse R1, R2 Rotation Matrix (UnQuantized)
  2. Convert to gguf file (UnQuantized)
  3. Quantize to data_types supported in llama.cpp (ex) Q8_0,Q4_0 (Quantized)
    However, Models fused by R1,R2 Matrix show worse perplexity than models quantized by Round-to-Nearest supported in llama.cpp.

In order to implement this, I've skipped fusing R4 Inverse Matrix,since it would require additional implementation to llama.cpp file.

Have you found these issues. If so, could you reveal the reason behind this?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions