https://huggingface.co/EleutherAI/gpt-j-6b
I'm almost certain that GPTJForCausalLM is not yet supported by llama.cpp nor will it likely ever be which is such a shame given how many amazing old GPT-J based models exist on HuggingFace. The main issue with GPT-J and similar very old models is that they predate llama.cpp itself so by the time llama.cpp got popular those models already lost interest and nobody really seems to be willing to spend thair spare time implementing support for old legacy models despite supporting them would be quite important for historical purposes.
My asumption was correct and it is indeed unfortinately not yet supported by llama.cpp:
INFO:hf-to-gguf:Loading model: gpt-j-6b
INFO:hf-to-gguf:Model architecture: GPTJForCausalLM
ERROR:hf-to-gguf:Model GPTJForCausalLM is not supported
By the way the same applies for OPTForCausalLM . An amazing legacy model which lost interest before llama.cpp took off.
I so agree. It's a real shame to not have those classic models, especially koboldai.