llama.cpp - GGUF support

by Doctor-Chad-PhD - opened Jun 6

Discussion

Doctor-Chad-PhD

Jun 6

Hi,

Is this model not supported by llama.cpp (gguf format)?
Or is there an error in the implementation?

I'm getting this message when trying to quantize to gguf at the moment:

  File "convert_hf_to_gguf.py", line 2553, in set_gguf_parameters
    logit_scale = self.hparams["hidden_size"] / self.hparams["dim_model_base"]
                                                ~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^
KeyError: 'dim_model_base'

Thank you for your time.

Doctor-Chad-PhD changed discussion title from GGUF support to llama.cpp - GGUF support Jun 6

Doctor-Chad-PhD

Jun 6

BigDong

OpenBMB org Jun 6

We have already fixed this error. add dim_model_base = 256. You can update the model and try again

Doctor-Chad-PhD

Jun 7

I don't understand why you didn't just merge my commit?
I created that commit myself, then your team member "xcjthu" copied it (6 hours later), and merged it as his own commit here.
Then you closed my original commit and said that it had already been fixed?

xcjthu

OpenBMB org Jun 10

I sincerely apologize for the oversight. When I was working on fixing this bug, I didn't notice your commit. I truly appreciate your contribution and help with our project.

You're absolutely right to point this out - we should have acknowledged your original work properly. We'll improve our review process to prevent this from happening again.

Doctor-Chad-PhD

Jun 10

No problem, thank you ❤️

xcjthu changed discussion status to closed Jun 11

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment