Misc. bug:convert_hf_to_gguf.py lora fine-tuning effects for visual recognition are completely lost – only the text-related fine-tuning remains effective.

[YuBin8]

Name and Version

version:b5952

Operating systems

Windows

Which llama.cpp modules do you know to be affected?

No response

Command line

Problem description & steps to reproduce

I specifically used this command to generate the mmproj file during conversion: python.exe convert_hf_to_gguf.py D:\AI\Model\TestModel --mmproj --outfile D:\AI\Model\GGUF\Test-7b-instruct.f16.gguf --outtype f16

Despite including the --mmproj flag, the visual capabilities are still missing in the mmproj GGUF model.

First Bad Commit

No response

Relevant log output