Transformers
falcon

error loading model: invalid character executing joined splited files

#12
by Althenwolf - opened

Hi every one. I have downloaded the different splits of the falcon-180b-chat.Q5_K_S.gguf-split-* model. Once downloaded I have joined them with:

cat falcon-180b-chat.Q5_K_S.gguf-split-* > /workvols/data2/mpperez3/projects/llama_cpp/llama.cpp/models/falcon-180b-chat.Q5_K_S.gguf

However when I try to use the model I get the error:

error loading model: invalid character
llama_load_model_from_file: failed to load model
llama_init_from_gpt_params: error: failed to load model '../../models/falcon-180b-chat.Q5_K_S.gguf'

I have tried with other models (without split) and they all work correctly with the same command. However with this one it is not possible. Does anyone know where the problem could be?

yeah, something with the new llama.cpp since a couple of days ago

Indeed, the problem is with llama.cpp main. I ran the model from python with the llama.cpp library and everything worked without problems. Many Thanks!

The files needs to be converted again: https://github.com/ggerganov/llama.cpp/issues/3484

i have converted it and am reuploading

EDIT: nvm. blokes on it.

Thanks for the report, I've triggered a re-build of the GGUF models, they'll re-upload over the next few hours

Re-made GGUFs have been uploading for the last 30 mins or so, and will finish in an hour or two. I've confirmed the new GGUFs work fine with latest llama.cpp.

I'll do Falcon 180B after that

Sign up or log in to comment