Thiago Machado
Thiago Machado
To add info on this, this is how I was able to define a unet: (Note: in this case I was using a version of dfdx that had some other...
I'll prioritize moving this experiment to a separate crate, but feel free to ping in case anyone have some question or suggestion.
I'm not sure if you got it working, but I'm trying to learn ML while using this lib and this is my dockerfile dev env: ```Dockerfile FROM nvidia/cuda:12.1.0-cudnn8-devel-ubuntu22.04 # basic...
I observed that on a 3x3 image, the center is also scaled incorrectly. But the distortion increases at the edge.
I'm not entirely sure, but I believe you can create a wrapper structure that defines how the `forward_mut` method behaves (assuming you want to implement a Module), and in that...
@kohya-ss Disclaimer: I have barely started to study all of this, so I apologize if I say something that doesn't make sense. From what I read so far, if we...
I've added a small PR with a temporary fix, which may indicate how the Cuda training (in general) is going wrong.
Could it be that the initial CudaCopyToDevice calls (made at the start of every iteration) are always overwriting the latest GPU weight values with the (static, initial) CPU weight values?
I'm re-opening because there is a simpler test, and also because if the PR is closed no new commits appear on it (which makes sense). The added test is a...
I added a check on src == dst during shape expansion, and made it stop the actual shape modification in case it is the same. But although the small test...