converting models not possible again

[DigitalRudeness]

Hi at all, the conversion of models isn't possible again.
The last time having the problem I did the mistake to not update the gguf python package. Now I'm on thelatest version (0.17.1) but getting a similar message:

EDIT: the path to the connversion file is a symlink. Just to metion

As the conversion is part of a scripted process including updating of the llama.cpp codebase / model *.safetensors etc, compiling the codebase and convertinng the models afterwards, commenting out the failing lines isn't an option because it will be overwritten the next time the process is triggered.

Thanks in advance for any kind of help :-)

[CISC]

Your problem is due to the symlink, if you instead run convert_hf_to_gguf.py directly from within the repository it will use the gguf-py contained therein instead of the older gguf package you have installed.

[DigitalRudeness]

This indeed did the trick! Master CISC is delivering the quick solution, exactly like in my last request... :-)

May I ask two questions just for my own curiosity?

1. what is the sense of this behavior that linked scripts are using the external gguf package but called at the absolute path it's using the built in variant?

2. Am I right in my assumption that I even don't need the external gguf package when I further follow this rule of calling the built in scripts?

[CISC]

1. what is the sense of this behavior that linked scripts are using the external gguf package but called at the absolute path it's using the built in variant?

The gguf package is imported like this:

llama.cpp/convert_hf_to_gguf.py

Lines 28 to 30 in f5e96b3

	if 'NO_LOCAL_GGUF' not in os.environ:
	sys.path.insert(1, str(Path(__file__).parent / 'gguf-py'))
	import gguf

That doesn't work when you symlink because the file path is then outside the repo.

2. Am I right in my assumption that I even don't need the external gguf package when I further follow this rule of calling the built in scripts?

Correct.

[DigitalRudeness]

I should read the code more carefully in the future... :-)

Many thanks to your patience and explanations! :-)