signalprime

Thank you Alex. I deleted my post after finding the location as well. My greatest compliments go to you because Sample Factory is truly a work of art. I find...

I just realized it requires CUDA even building without

that's because the optimizer was not specified during checkpoint creation, it is not critical for inference

> For MacOS, we have to use the C++ implementation : https://github.com/ggerganov/llama.cpp > > Works like a charm on my side, with the 3 models that fit in my RAM...

I'm digging into the C++ for it https://github.com/pytorch/pytorch/blob/834c7a1d3ea07878ad87d127ee28606fc140b552/aten/src/ATen/native/WeightNorm.cpp#L50 I'm fine with C++ but not configured to build the MLX project.. questioning motivation on a Saturday night haha. Do we have...

That's great, and thanks a lot for the input @awni. Part of me was thinking that too.

any idea where this is happening in the code? is it a bad NPM or is it something in this repository?

for real?

hey crew, I spent a few hours last night and today working on both CPU and MPS updates to this codebase. I also ran into the same results as @akorzh...

> This is for sure one of those researches that I'm sceptic about. The animation is just TO good to be img2video from audio. Muscles are moving, backgrounds are visible...