fun hack, GPT-2 inference running on my jailbroken Kindle Scribe
this is real speed, like ~13 tok/s at 4-bit quant of the 124M sized model