Matthew Bryant

The code is heavily inspired by the NinePatch implementation so I hope it's not too much effort! The only big style concern I had (which I realized after finishing the...

I'm currently on vacation so I can't review this at the moment. This repo was more of my test repo for the tutorial project and there is another one with...

Thank you, I appreciate the commits and I'll leave this open so others can see how to upgrade to 0.8. However due to how tightly coupled this code base is...

I believe I resolved them in the github editor. It was just adding the files and use statements to mod.

As someone who makes content around Bevy (LogicProjects Youtube), the reddit was an important place for me to get the first 100 eyes on my work. I post less there...

I'm seeing the same original problem with the recommended drivers, hangs after closing without panic. ```2024-02-19T20:41:39.728417Z INFO bevy_winit::system: Creating new window "App" (0v1) 2024-02-19T20:41:39.728702Z INFO log: Guessed window scale factor:...

Hmm with `python server.py --listen --no-stream --auto-devices --load-in-8bit` gpu memory increases until it crashes with the normal OOM error. With `python server.py --listen --no-stream --gpu-memory 1 --auto-devices --load-in-8bit` I get...

Sorry I was unclear, I tried every single number and all fail the same way, even 1 which should have the easiest time working if I'm understanding correctly.

I have the same behaviour, --gpu-memory 7 goes through the whole loading process and then crashes with OOM. Anything less crashes instantly with the error message from the start of...

I have the same error, happens on https://github.com/setzer22/llama-rs/blob/3ce15b1200a3419d31c2dbe44b4ebd569370409a/llama-rs/src/llama.rs#L446 with tensor.nelements() = 212992000, n_parts = 3, and nelementes = 53248000 llama.cpp returns the same values with n_parts = 4 instead so...