dusty jasper
I don't really know what this error is pointing to or how to resolve it. It seems that it's fairly unique to me as I'm not seeing it come up on google or discord searches. Anyone able to point me in the right direction of how to diagnose and solve this?
I'm trying it with multiple ggml files (I'm cpu only). Strange thing is most of these models work fine with koboldcpp. Any ideas?
viscid hare
I think llama.cpp had a major update recently that changed how the model files it can use are structured. All new ggml files (from the last few days) are in this new format. I don't know that the llama-cpp-python implementation that Ooba uses can read those new files just yet. That might be why you're seeing those messages.
Maybe try a ggml file from before 2-3 days ago and see if it works a bit better?
One of these models might work as a test: https://huggingface.co/TheBloke/gpt4-x-vicuna-13B-GGML/tree/previous_llama