tcmaps

` install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia -c conda-forge -y` ``` real 3m5.016s user 2m53.217s sys 0m3.843s (conda) studio-lab-user@default:~$ ``` ``` real 0m9.198s user 0m16.509s sys 0m1.793s...

> ```diff > dependencies: > - pip: > - - opencv-contrib-python==4.3.0.36 > + - opencv-contrib-python > ``` Take the chance to also replace opencv-contrib-python with opencv-contrib-python-headless, it skips some nasty...

But does it really have to be integrated? Set up ngrok are 2 commands, and 1 to run it.

I used ngrok on sagemaker and hadn't anything to complain so far, seems even faster than cloudflared. Only caveat may be the session limit of 1h with a free plan....

oh actually my initial question was more directed at op, whether integration/extension into the repo are even required, because it's simple enough to use as standalone package.

(lib)mamba support would be awesome

Here I adapted the sdxl diffusers example to use both T4 on kaggle: ```python from diffusers import DiffusionPipeline import torch import time torch.cuda.reset_peak_memory_stats() base = DiffusionPipeline.from_pretrained( "stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16, variant="fp16", use_safetensors=True...