Show HN: Open-source alternative to AWS BlazingText, in PyTorch

Category: ai-ml

Tags: word-embeddings, fasttext, gpu-training, pytorch, nlp

Score: 6.5/10 (Innovation: 6, Technical: 7, Documentation: 8, Utility: 5)

floret-torch reimplements fastText/floret word vector training in PyTorch to leverage GPU memory bandwidth, achieving ~20x speedup over CPU training while maintaining hash parity and vector quality. It provides a practical open-source alternative to AWS BlazingText for training subword embeddings on your own GPU.

Target audience: NLP researchers and ML engineers

Repository: https://github.com/Fazel94/floret-torch · Python · MIT · 7 stars

View on Hacker News