drhead
drhead
## Description * Implements an option to skip CFG on early sampling steps as recommended in https://arxiv.org/abs/2404.07724 * Somewhat awkward to use as it stands. ~~I'll probably change it to...
### Current Behavior I have a project running where I am storing a Plotly graph over time, which has 1000 points and which I generate/store every 100 training steps. Viewing...
Contract Configurator currently has a HasAntenna parameter which allows contracts to require antennas of a specific power level and relay status, but that can't be checked and is also fundamentally...
## Description I have vastly improved the performance of live preview by making two changes: * running the decode on its own CUDA stream which lets operations parallelize * making...
## Description * Currently, the k diffusion sampler code creates a tensor for sigmas on the GPU. This will cause forced device syncs every step because they are used for...
## Description Change 1: Timestep Embedding Patch * Fixes a blocking op in the timestep embedding. It was creating a tensor on CPU and then moving it to GPU, which...
### Feature request It would be extremely helpful to have stochastic rounding support, as described in [Revisiting BFloat16 Training](https://arxiv.org/pdf/2010.06192.pdf), for 8-bit optimizers. This would require working with a full-precision set...
I would greatly appreciate it if this plugin had support for zero terminal SNR models. ComfyUI already has good support for these models, and supporting them in this would be...
This fixes a shape mismatch with controlnet that at least exists with the reference_only processor. Possibly others too, but I haven't tested and this shouldn't wrongly affect anything except similar...
This would close #108. I changed the one operation that actually relies on sigmas being on the same device as the model samples and changed it so that it is...