4 ms·
Probably, you have to load the weights from disk at some point.
by fassssst 3y ago
Probably, you have to load the weights from disk at some point.
- echelon 3y agoThat's exactly it. These models are huge.
- cal85 3y agoI’m probably missing something but if the bottleneck is disk read speed, wouldn’t it only take about 5-6 seconds to fill the entire 32GB memory from disk? I just googled and found a benchmark quoting 5,507 MB/s read on an M1 Max.
- liuliu 3y agoPyTorch checkpoint is slow to load.
- brucethemoose2 3y agoThe diffusers format this repo uses should be faster, but there is still some overhead, yeah.
- Yoric 3y agoYeah, the PyTorch disk format is pretty bad.