shady VS Cgml

Sounds cool, but this requires yet another language to learn[0]. As someone who only has limited knowledge in this space, could someone tell me how comparable is the compute functionality of rust-gpu[1], where I can just write rust?

[0] https://github.com/Hugobros3/shady#language-syntax

[1] https://github.com/EmbarkStudios/rust-gpu

> If there is a future with this idea, its gotta be just shipping the LLM with game right?

That might be a nice application for this library of mine: https://github.com/Const-me/Cgml/

That’s an open source Mistral ML model implementation which runs on GPUs (all of them, not just nVidia), takes 4.5GB on disk, uses under 6GB of VRAM, and optimized for interactive single-user use case. Probably fast enough for that application.

You wouldn’t want in-game dialogues with the original model though. Game developers would need to finetune, retrain and/or do something else with these weights and/or my implementation.

If you just want to run Mistral on Windows, you could try my port: https://github.com/Const-me/Cgml/tree/master/Mistral/Mistral...

The setup is relatively easy: install .NET runtime, download 4.5 GB model file from BitTorrent, unpack a small ZIP file and run the EXE.

Speaking about random sampling during inference, most ML models are doing it rather inefficiently.

Here’s a better way: https://github.com/Const-me/Cgml/blob/master/Readme.md#rando...

My HLSL is easily portable to CUDA, which has `__syncthreads` and `atomicInc` intrinsics.

I did a few times with Direct3D 11 compute shaders. Here’s an open-source example: https://github.com/Const-me/Cgml

Pretty sure Vulkan gonna work equally well, at the very least there’s an open source DXVK project which implements D3D11 on top of Vulkan.

Here’s an example of a custom 4 bits/weight codec for ML weights:

https://github.com/Const-me/Cgml/blob/master/Readme.md#bcml1...

llama.cpp does it slightly differently but still, AFAIK their quantized data formats are conceptually similar to my codec.

> the API was high-friction due to the shader language, and the glue between shader and CPU

Direct3D 11 compute shaders share these things with Vulkan, yet D3D11 is relatively easy to use. For example, see that library which implements ML-targeted compute shaders for C# with minimal friction: https://github.com/Const-me/Cgml The backend implemented in C++ is rather simple, just binds resources and dispatches these shaders.

I think the main usability issue with Vulkan is API design. Vulkan was only designed with AAA game engines in mind. The developers of these game engines have borderline unlimited budgets, and their requirements are very different from ordinary folks who want to leverage GPU hardware.

Minor update https://github.com/Const-me/Cgml/releases/tag/1.1a Can’t edit that comment anymore, too late.

For implementing stuff from scratch, if you use Windows you could try my C# based library for that: https://github.com/Const-me/Cgml/

It’s vendor agnostic, so HLSL instead of CUDA or Triton. Here’s the compute shaders implementing inference of Mistral-7B model: https://github.com/Const-me/Cgml/tree/master/Mistral/Mistral...