finalfusion-rust and finalfrontier
Finalfrontier is a training tool that generates word embedding models, while finalfusion-rust is the inference runtime that loads and uses those trained models, making them complements in a producer-consumer relationship.
About finalfusion-rust
finalfusion/finalfusion-rust
finalfusion embeddings in Rust
Supports multiple vocabulary and storage backends including memory-mapped and product-quantized embeddings for flexible performance/memory tradeoffs, plus cross-format conversion (fastText, word2vec, GloVe). Enables similarity and analogy queries through `ndarray`-backed vector operations, with optional BLAS/LAPACK acceleration for quantized lookups. Integrates with the finalfusion ecosystem for training embeddings via finalfrontier and quantization via the reductive crate.
About finalfrontier
finalfusion/finalfrontier
Context-sensitive word embeddings with subwords. In Rust.
Supports multiple training architectures (skip-gram variants, dependency-based models) with noise contrastive estimation and Hogwild! parallelization for efficient training. Embeddings export to multiple formats including the native finalfusion format, fastText, word2vec, and GloVe, with downstream quantization support via the finalfusion utilities. Integrates with the finalfusion Rust crate and Python module for inference and embedding manipulation.
Related comparisons
Scores updated daily from GitHub, PyPI, and npm data. How scores work