convert-llama-ggmlv3-to-gguf: Try to handle files older than GGJTv3

[KerfuffleV2]

• Better error messages for files that cannot be converted

• Add file type to GGUF output

Closes #2990

[KerfuffleV2]

I'll hold off merging this for a bit until I get some feedback from the people with older GGML files.

[cebtenzzre]

Isn't the name of the script technically inaccurate now? Maybe we should at least add something to --help to indicate to users that certain pre-GGMLv3 files are supported as well.

[KerfuffleV2]

Isn't the name of the script technically inaccurate now?

Not just technically. :) I kind of wanted to just rename it, but I felt like that would be pretty confusing for existing users.

Maybe we should at least add something to --help

Sure, that's not a bad idea. It seems like for the most part, users don't really know the exact type of file they have anyway and will just try to run the script but some additional information certainly can't hurt.

[KerfuffleV2]

You know what, I think I will just rename the file since ggmlv3 is wrong to start with. I don't know what I was thinking.

[cebtenzzre]

Lines of code should be at most 120 chars wide, for readability:

Suggested change

	if (self.file_format < GGMLFormat.GGJT or self.format_version < 2) and ftype not in (GGMLFType.ALL_F32, GGMLFType.MOSTLY_F16):
	raise ValueError(f'Quantizations changed in GGJTv2. Can only convert unquantized GGML files older than GGJTv2. Sorry, your {self.file_format.name}v{self.format_version} file of type {ftype.name} is not eligible for conversion.')
	if (self.file_format == GGMLFormat.GGJT and self.format_version == 2) and ftype in (GGMLFType.MOSTLY_Q4_0, GGMLFType.MOSTLY_Q4_1, GGMLFType.MOSTLY_Q4_1_SOME_F16, GGMLFType.MOSTLY_Q8_0):
	raise ValueError(f'Q4 and Q8 quantizations changed in GGJTv3. Sorry, your {self.file_format.name}v{self.format_version} file of type {ftype.name} is not eligible for conversion.')
	if ($
	(self.file_format < GGMLFormat.GGJT or self.format_version < 2)$
	and ftype not in (GGMLFType.ALL_F32, GGMLFType.MOSTLY_F16)$
	):$
	raise ValueError($
	'Quantizations changed in GGJTv2. Can only convert unquantized GGML files older than GGJTv2. Sorry, '$
	f'your {self.file_format.name}v{self.format_version} file of type {ftype.name} is not eligible for '$
	'conversion.'$
	)$
	if ($
	self.file_format == GGMLFormat.GGJT and self.format_version == 2$
	and ftype in ($
	GGMLFType.MOSTLY_Q4_0, GGMLFType.MOSTLY_Q4_1, GGMLFType.MOSTLY_Q4_1_SOME_F16, GGMLFType.MOSTLY_Q8_0,$
	)$
	):$
	raise ValueError($
	f'Q4 and Q8 quantizations changed in GGJTv3. Sorry, your {self.file_format.name}v{self.format_version} '$
	f'file of type {ftype.name} is not eligible for conversion.'$
	)

[KerfuffleV2]

Thanks for the feedback.

In this particular case, I'm honestly not sure putting the time into improving the formatting is really worth it. I doubt anyone else is going to be messing with this script, and it's probably not even going to be in the repo in a month or so. But I made some formatting cleanups anyway.