How facebookresearch/fasttext works
fastText is a library and CLI for learning word representations and text classifiers, then querying or evaluating the resulting models. The main documented path is a user command routed by the CLI into the FastText engine, which trains or loads a model and produces vectors or predictions. Separate branches cover Python access, model compression, and bilingual embedding alignment. The README establishes the main use cases; edges from sampled code are limited to visible dispatch and module references, so unsampled internal call paths are omitted.
Main entry points: fasttext.cc FastText engine · main.cc CLI commands · utils.py Alignment utilities · FastText.py Python API
Components 21
Core engine
- Arguments
src/args.cc - FastText engine
src/fasttext.cc - Dictionary
src/dictionary.cc - Training model
src/model.cc - Training losses
src/loss.cc - Vector matrices
src/densematrix.cc - Model quantization
src/quantmatrix.cc
User interfaces
- CLI commands
src/main.cc - Python API
python/fasttext_module/fasttext/FastText.py - Native binding
python/fasttext_module/fasttext/pybind/fasttext_pybind.cc - WebAssembly API
webassembly/fasttext_wasm.cc - JavaScript wrapper
webassembly/fasttext.js
Embedding alignment
- Supervised alignment
alignment/align.py - Unsupervised alignment
alignment/unsup_align.py - Multilingual alignment
alignment/unsup_multialign.py - Alignment utilities
alignment/utils.py
Model utilities
- Python utilities
python/fasttext_module/fasttext/util/util.py
Other components
- User
- Model files
- Training text
- Embedding files
Connections 17
- User → CLI commands: runs command
- CLI commands → FastText engine: dispatches
src/main.cc - CLI commands → Arguments: parses options
- Training text → FastText engine: trains from
- FastText engine → Model files: saves or loads
- FastText engine → Embedding files: exports vectors
- FastText engine → Dictionary: uses vocabulary
- FastText engine → Training model: optimizes model
- FastText engine → Training losses: creates loss
- FastText engine → Vector matrices: uses matrices
- FastText engine → Model quantization: quantizes model
- Python API → Native binding: calls native API
- JavaScript wrapper → WebAssembly API: calls bindings
- Supervised alignment → Alignment utilities: loads and evaluates
alignment/align.py - Unsupervised alignment → Alignment utilities: uses helpers
alignment/unsup_align.py - Multilingual alignment → Alignment utilities: uses helpers
alignment/unsup_multialign.py - Python API → Python utilities: offers utilities