A 4-bit, 128g, act_order=True GPTQ quantisation of JackFram/llama-68m, a 68 million parameter Llama1 model; created on request for software testing.

Not for normal usage!