ask Replicate staff to help us download the user's Safetensor model with pget() (which is paralleL) instead of regular download when using upscaler because now it takes 75 seconds to download and that costs us about $12,000/mo in GPU time billed #photoai