bennyschmidt/next-token-prediction

Supports both n-gram and optional vector embedding-based prediction with 144-dimensional token pair embeddings for semantic context, switchable via environment variable. Trains directly on local .txt files or pre-loaded datasets without external dependencies, generating completion suggestions, sequence predictions, and similarity-based alternatives. Includes readline and browser UI examples for autocomplete, search, and spell-checking applications.

143 stars. No commits in the last 6 months.