A collection of tools for building and training Transformers using tch-rs (a Rust bindings for C++ PyTorch API).
The best way to explore it is to start from running an example
Assuming an NVIDIA GPU is available, after cloning this repository, an example can be lauched with these shell commands:
export TORCH_CUDA_VERSION=cu118
cargo run --example run-albert --features=examples
Once launched, the example will
- Download the WikiText dataset.
- Parse its training/validation/test documents and train a tokenizer on them.
- Tokenize the documents with the tokenizer.
- Launch training of an ALBERT model using MLM task.
- Keep the downloaded dataset, the trained tokenizer, and the model's weights inside a "cache" subdirectory, to facilitate restarts.