smarty_pants VS smolrsrwkv

The idea is that you can simply use this crate with your project to easily train a neural network using your project. The library supports creating, training, parsing, and running. It may gain more functionality in the future. As it stands it's quite small and pretty fast with 5 NeuralNetworks taking nano-seconds to train 1000 generations in the example program. I've tried to make sure that it is "complete" and as such, I've documented nearly every function, method, and struct. I've also written an example project and tried to make it relatively easy to use.

For an example of actually using it, you can look at my little project for running inference on RWKV models: https://github.com/KerfuffleV2/smolrsrwkv It uses repugnant-pickle to enable loading PyTorch models directly with no conversion requirement or Python dependencies.

If you want some proof, I wrote my own application that can run inference on RWKV models (competing approach similar to GPT which most LLMs use currently): https://github.com/KerfuffleV2/smolrsrwkv

It can now quantize to 8bit for 4x memory savings: https://github.com/KerfuffleV2/smolrsrwkv/tree/experiment-quantize

Here's an implementation of RWKV I wrote that can run inference on models: https://github.com/KerfuffleV2/smolrsrwkv

Now, I'm not an expert but I do know a little more than the average person. I actually just got done implementing a simple one based on the RWKV approach rather than transformers: https://github.com/KerfuffleV2/smolrsrwkv