afgonczol
afgonczol
I get the same error. It seems to happen when using pre_layer > 0.
As Ph0rkk0z mentioned, it's due to a mismatch between the gptq used with oobabooga and the gptq used to quantize the model. I've found that 4bit maintains versions of the...
> Looks like it's caused by using pre_layer. No solution yet that I'm aware of.
> Get extra RAM (at last 32GB) and use the 30B model. It provides quite long replies and has excellent reasoning skills. Can you share what settings and prompts you're...