convert : allow quantizing lora again

[CISC]

Allow quantizing LoRA at conversion again, but default to F32 (as has been the norm since #8980 inadvertently forced this).

Fixes #17447
Fixes #10671

[ggerganov]

Did this change somehow cause some of the CPU CI to start failing:

[ggerganov]

@CISC For some reason this exception is not being caught when running in the CI:

https://github.com/ggml-org/llama.cpp/actions/runs/19645856267/job/56260791630#step:5:6742

Locally it is being caught and the execution proceeds normally. Any ideas?

[CISC]

Did this change somehow cause some of the CPU CI to start failing:

I don't see how, must be something else.

@CISC For some reason this exception is not being caught when running in the CI:

https://github.com/ggml-org/llama.cpp/actions/runs/19645856267/job/56260791630#step:5:6742

Locally it is being caught and the execution proceeds normally. Any ideas?

Strange, I'll have a look...

[ggerganov]

I traced it that transformers==4.57.2 breaks the conversion and throws the second exception in the log. Downgrading to transformers==4.57.1 resolves the problem.

[CISC]

I traced it that transformers==4.57.2 breaks the conversion and throws the second exception in the log. Downgrading to transformers==4.57.1 resolves the problem.

Yeah, I just found that out too: huggingface/transformers#42378

[CISC]

There will be a hotfix as soon as huggingface/transformers#42389 merges.