fBLT

Fast Byte Latent Transformer in pure C and CUDA. A tokenizer-free, byte-level language model for code completion and small LLMs, based on Meta's BLT and Fast BLT papers.

GitHub

Official source

Start locally with the CLI guide