Awesome GitHub RepositoriesTokenizers

This project is a speech recognition and translation engine that utilizes a sequence-to-sequence transformer architecture to convert audio into text. It is built upon a weakly supervised learning framework, which leverages large-scale, unlabelled audio-transcript data to create generalized speech representations capable of performing simultaneous transcription, language identification, and translation. The system distinguishes itself through a unified multi-task modeling approach that shares token sequences across different objectives, allowing it to handle diverse languages and vocabularies

Codex is an automated programming tool and generative code assistant designed to interpret developer intent through a natural language interface. It functions as a machine learning model trained on public code repositories to provide intelligent code completion, suggestions, and refactoring within development environments. By translating human instructions into executable code snippets, the system bridges the gap between high-level technical requirements and functional software implementation. The engine utilizes transformer-based sequence modeling and supervised fine-tuning to align its outp

nanoGPT is a lightweight engine for training and fine-tuning transformer-based language models from scratch. It provides a minimalist codebase designed for educational exploration and rapid experimentation with neural network architectures, utilizing self-attention and feed-forward layers to process sequences and predict subsequent elements. The project distinguishes itself through a focus on high-speed data ingestion and hardware-accelerated performance. It includes a dedicated pipeline for transforming raw text into memory-mapped binary files, which enables efficient streaming during traini

This project provides a comprehensive framework for building, training, and managing autonomous agents. It enables the construction of systems that utilize language models to plan, manage memory, and execute multi-step tasks through iterative reasoning loops and tool-based actions. The framework distinguishes itself by offering specialized capabilities for interacting with graphical user interfaces and legacy software, allowing agents to perceive visual elements and perform actions like a human user. It supports complex, cross-application workflows through graph-based orchestration and provid

Llama is a computational framework and runtime environment designed for executing transformer-based neural networks locally. It functions as a generative AI inference engine, enabling the processing of input sequences through pre-trained model weights to produce text completions and structured data outputs directly on your own hardware. The system distinguishes itself through specialized memory and computation management techniques, including memory-mapped weight loading and quantization-aware inference, which allow for efficient execution on standard consumer hardware. It utilizes a stateles

Meilisearch is a Rust-based search engine providing typo-tolerant full-text and vector-based semantic search with real-time conversational capabilities.

spaCy is a Python natural language processing framework designed for industrial-scale text processing. It converts raw text into structured data for machine learning pipelines through a combination of statistical language model trainers, transformer-based text processors, and syntactic dependency parsers. The project enables the integration of pretrained transformer architectures to perform complex linguistic analysis and multi-task learning. It also provides a specialized system for neural named entity recognition to identify and categorize key entities within text. The framework covers a b

This project is an educational platform and research toolkit designed to teach deep learning through a combination of mathematical theory, visual diagrams, and executable code. It provides a comprehensive environment for building, training, and evaluating neural networks, grounding complex concepts in interactive computational notebooks that allow for hands-on experimentation. The framework distinguishes itself by interleaving theoretical foundations—including linear algebra, calculus, and probability—with practical implementations across multiple industry-standard libraries. It supports flex

Guidance is a generative AI orchestration framework designed to manage complex interactions with language models by embedding programmatic control directly into the prompt generation process. It functions as a prompt programming environment that allows developers to interleave raw text with executable logic, enabling the construction of sophisticated, multi-step agentic workflows. The framework distinguishes itself through grammar-constrained token sampling and stateful stream interception, which restrict the model's output distribution based on formal language rules. By enforcing these const

Sonic is a high-performance, lightweight search backend designed to provide real-time full-text search and autocomplete capabilities for applications. It functions as a persistent indexing server that maps text terms to object identifiers, allowing developers to integrate rapid search functionality without storing raw document content directly within the search engine. The system distinguishes itself through a specialized graph-based index that enables real-time word prediction and typo correction. Communication is handled via a custom, low-latency binary protocol over raw TCP sockets, which

Flair is a transformer-based natural language processing framework used to build and train models for text classification and sequence tagging. It provides a specialized library for generating contextual text embeddings and performing linguistic analysis. The framework includes dedicated tools for named entity recognition, including the identification of specialized biomedical entities across multiple languages. It further supports entity linking to map identified text mentions to unique entries within general or biomedical knowledge bases. The project covers a broad range of language analys

Outlines is a guided text generation framework and structured output engine for large language models. It enforces precise structural constraints on model output during the sampling process to ensure the generation of valid data. The framework ensures that model outputs strictly adhere to predefined data models, including JSON schemas, regular expressions, and formal grammars. This enables the conversion of natural language inputs into structured arguments for function calling and the generation of valid JSON for downstream processing. The system manages model orchestration through prompt te

Outlines is a library designed to ensure machine-readable output from generative models by applying programmatic constraints during the token sampling process. It functions as a toolkit for forcing large language models to generate text that strictly adheres to JSON schemas, regular expressions, and formal grammars, enabling the integration of model responses into existing software systems. The library distinguishes itself by integrating formal language rules directly into the sampling loop. It achieves this by converting regular expressions into deterministic finite automata and utilizing lo

SentencePiece is a text segmentation engine and tokenization library designed for machine learning workflows. It provides a comprehensive toolkit for transforming raw text into subword units or numerical identifiers, enabling consistent data representation for neural network training and inference. The library supports the training of segmentation models from raw text, allowing for the creation of custom vocabularies tailored to specific domain requirements. The project distinguishes itself through its byte-level encoding and fallback mechanisms, which ensure that every input can be represent

Minimal, clean code for the Byte Pair Encoding (BPE) algorithm commonly used in LLM tokenization.

LLMLingua is a prompt compression tool that reduces token count in prompts before they are sent to a large language model, cutting API costs and latency while preserving task performance. It operates as an extractive pipeline using a BERT-level Transformer encoder to classify each token for removal based on full bidirectional context from the prompt, retaining only key information and discarding non-essential tokens. The tool is trained through a knowledge distillation process, where a compact compression model learns from an extractive dataset derived from a large language model's output to

Orange3 is a visual data mining platform that provides an interactive canvas for building data analysis workflows without writing code. At its core, it offers a widget-based visual programming environment where users connect configurable components to perform data preprocessing, machine learning model training, statistical evaluation, and interactive visualization. The platform is built on NumPy-backed data tables with domain descriptors that define variable names, types, and roles, and includes a lazy SQL query proxy for working with database tables without loading all data into memory. The

onnxsim est un optimiseur de graphes de deep learning et un simplificateur de modèles conçu pour réduire la complexité des graphes de calcul ONNX. Il fonctionne comme un compresseur de modèle qui remplace des séquences d'opérateurs complexes par des sorties constantes simplifiées afin de réduire la surcharge opérationnelle. Le projet réalise cette simplification grâce à l'inférence par pliage de constantes (constant folding), qui remplace les sous-graphes d'opérateurs constants par des tenseurs constants pré-calculés. Il utilise la réécriture de graphes basée sur des motifs et l'analyse statique de graphes de calcul pour identifier et supprimer les nœuds redondants ou les opérations inaccessibles. L'outil couvre de larges capacités d'optimisation de modèles, notamment l'élimination de la redondance des opérateurs et la suppression des nœuds de reshape ou d'identité inutiles. Ces processus rationalisent le flux d'exécution et réduisent l'empreinte mémoire du modèle.

Mapshaper est un outil pour traiter, simplifier et convertir des données vectorielles géographiques, disponible sous forme d'interface en ligne de commande, d'outil de navigateur web et de bibliothèque Node.js. Il fonctionne comme un projecteur de coordonnées, un convertisseur de données vectorielles et un optimiseur d'actifs de carte web conçu pour transformer les jeux de données spatiaux entre différents systèmes de référence de coordonnées et formats de fichiers. Le projet se distingue par sa simplification de géométrie préservant la topologie, qui réduit le nombre de sommets tout en maintenant les limites partagées pour éviter les lacunes et les chevauchements. Il optimise davantage les actifs pour le web grâce à la quantification des coordonnées et au filtrage des attributs pour réduire la taille des fichiers. Le système couvre un large éventail de capacités, y compris la reprojection de coordonnées utilisant des chaînes PROJ et des codes EPSG, et la conversion de données entre des formats tels que Shapefile, GeoJSON, TopoJSON, GeoPackage et KML. Il fournit des outils de traitement de géométrie étendus pour la mise en mémoire tampon, le découpage, la dissolution et la réparation des topologies, ainsi que des utilitaires de gestion de données pour la jointure, le filtrage et la transformation d'attributs. De plus, il inclut des fonctionnalités de visualisation pour générer des exportations SVG stylisées, des graticules et des cartes à symboles proportionnels. Les capacités de traitement spatial peuvent être intégrées directement dans les applications JavaScript et les pipelines de build via sa bibliothèque Node.js.

Pretrained-Language-Model is a machine learning library and natural language processing toolkit designed for pretraining, tokenizing, and compressing large language models using transformer architectures and specialized optimization techniques. It supports Chinese and multilingual natural language processing tasks, including text classification and conversational response generation. The framework provides specialized capabilities for training large-scale autoregressive and contextual language models, alongside model compression techniques like knowledge distillation and quantization to reduc