Rust ML IG Meeting 00039

Agenda

enzyme integration into rust compiler
- present work at the EnzymeConf23
- demonstrate a simple example
- typetrees causing issues at the moment (no complex compound types)
progress of the rust-ml/book
- available at https://rust-ml.github.io/book/
- simplified structure (removed special backend module)
- anything in vision, image classification, transformers etc. would be interesting
- chris: linfa in use for image clustering (but not public yet)
- chris: writing a chapter about pure Rust introduction into machine learning would be interesting
- classification on cifar10 dataset first target
rust-ml/tsap crate
- copy behaviour of hydra
- take a look at DVC, for large scale pipelines https://towardsdev.com/ml-pipeline-decoupled-i-managed-to-write-a-framework-agnostic-ml-pipeline-with-dvc-rust-python-287de68104c9
GPU support
- Corey works on wrapper for CUDA toolkit with JIT compilation of cuda kernels https://github.com/coreylowman/cudarc
- rust-cuda project has not got updates recently but integration into nalgebra in progress
- Burn uses torch as GPU backend
- chris: wgpu also supports computation kernels but apparently no collaboration with embark group
- SPIRT https://github.com/EmbarkStudios/spirt
Funding: NumFocus https://numfocus.org/