This sounds like it uses the same techniques as the ones described in the "LLM in a Flash" paper posted yesterday? If so, cool to see an implementation of these

This sounds like it uses the same techniques as the ones described in the "LLM in a Flash" paper posted yesterday? If so, cool to see an implementation of these techniques running models on non-Apple GPUs.