4 ms·
but that’s not much useful if the model is nano. Webgpu would be a better point of misuse maybe
by asadalt 2y ago
but that’s not much useful if the model is nano. Webgpu would be a better point of misuse maybe
- btown 2y agoThe latency to download all the weights would be a limiting factor on WebGPU. But if the weights are already downloaded and optimized locally…