Amol Kapoor

Update: it looks like the PyPi located here (https://pypi.org/project/rasterfairy/) is not up to date. Instead, I used the PyPi for rasterfairy-py3 (https://pypi.org/project/rasterfairy-py3/) which is up to date. Leaving this open...

I know this issue hasn't been touched in a while, but I wanted to add that the stochastic nature of umap can come up even in not-so-boundary points. In my...

Digging into this a bit more, I managed to get this down to the following, even more minimal, example of hanging using clip: ``` import os import urllib from tqdm...

So as far as I can tell, in the second code block the child process is not referencing anything in the parent process. And I'm not using cuda, all of...

Got it down to an even simpler version than before: ``` import torch import clip from PIL import Image import multiprocessing as mp model = clip.model.CLIP(512, 224, 12, 768, 32,...

If I understand right, changing the context to `spawn` causes the model to reload, which sorta defeats the point of loading the model in the parent. In that scenario I...

``` ERROR: Could not find a version that satisfies the requirement fast-colorthief==0.0.4 (from versions: 0.0.1) ERROR: No matching distribution found for fast-colorthief==0.0.4 ``` On ubuntu, python 3.8.10

I got around this by running `pip install git+https://github.com/bedapisl/fast-colorthief.git#egg=fast-colorthief` obviously not a fix for the bug but may help some other folks who are running into issues.

hey there! Any progress on this one? I've been able to get around this limitation using casting for now, but just ran into a bug where I was casting incorrectly...

Just checking on this one again, somehow I'd thought this was resolved so was confused when I ran into this in a different place :sweat_smile: