ice-nine

simple ml library written in rust i made for fun/learning, heavily inspired by PyTorch's design

Usage example in ./examples/train_llm/main.rs:

cargo run -r --example train_llm

currently working

• basic tensor ops (naive CPU implementations)
• autograd/backprop
• really slow, tiny transformer can learn a toy distribution

todo

soon

• change tensor storage to use strides/support views
• model serialization (saving/loading)
• easy inference code for autoregressive generation

who knows

• optimize tensor ops
• gpu support?? (https://github.com/Rust-GPU/Rust-CUDA)