Hayden Prairie

@johnchienbronci, did you ever fix this issue, I am running into it too

DDIM allows for faster diffusion. It looks like they are taking 1000 diffusion steps total with 50 sampling steps, thus a stride of 20.

@karan6181 Is there an example of using streaming with composer, I'm currently struggling to get streaming working with multi-node processes, the composer documentation states that it is compatible, but doesn't...

@tridao I am curious if you had a reason for why it can work fine without the conv. From my understanding without the conv there would be no feature mixing...

Ahh nvm, I see that they could also come into play through the projection of B, C, and delta. However, when you say worse quality do you mean just worse...

@EricLina I might be a little late to this, but I was looking at something similar. The forward scan is used to recalculate hidden states, while the reverse scan is...

This looks like you are loading in a new value for $exp(delta, A)$ but then loading in the hidden state as 0. You should probably instead loading it in as:...

Yeah looks like a conda issue. Here is the stack trace. ```bash GNU gdb (Ubuntu 12.1-0ubuntu1~22.04) 12.1 Copyright (C) 2022 Free Software Foundation, Inc. License GPLv3+: GNU GPL version 3...

@nimlgen I have been looking through the code base to try to see how it is generating the source of "/usr/local/src/conda/python-3.11.8/Modules/_ctypes/cfield.c", but have been unsuccessful. Any ideas on where to...

kernargs isn't zero for me. I have a dedicated GPU, 7900 XTX. ```bash (tiny) prairie@TRX40:~/Projects$ DEBUG=3 python3 -c "from tinygrad import Tensor; N = 1024; a, b = Tensor.rand(N, N),...