2 ms·
that's openCL compute, LLM models ideally should be hitting the neural accelerator, not running on generalized gpu compute shaders.
by paulmd 3y ago
that's openCL compute, LLM models ideally should be hitting the neural accelerator, not running on generalized gpu compute shaders.