Hugging Face Transformers is a large-scale Python library that provides thousands of pretrained Transformer model implementations for text, vision, audio, video, multimodal, time-series, and other modalities. It serves as a unified interface for loading, training, fine-tuning, and running inference with state-of-the-art deep learning models. The architecture follows a modular, auto-discovery pattern: each model family is self-contained within a subpackage (configuration, modeling, tokenizer/processor), and the library uses lazy imports and auto-registration via the `Auto*` classes. Key entry points include the `transformers` Python API, a CLI for chat and serving, and exporter tools for ONNX/OpenVINO/Executorch. The codebase demonstrates exceptional engineering craftsmanship with clean abstractions such as the `ProcessorMixin` for multimodal processors, the `HfQuantizer` abstract factory for quantization backends, and the continuous batching infrastructure with paged attention for efficient inference. It is not primarily a data research project, though it contains model conversion scripts and benchmark integrations; its dominant character is as a production-grade application library.
Docling is a production-grade Python library for converting diverse document formats (PDF, DOCX, PPTX, XLSX, HTML, LaTeX, EPUB, images, audio, video, email, EBCDIC, and more) into a unified, structured `DoclingDocument` representation. It is designed to serve as the document understanding layer for AI workflows, particularly Retrieval-Augmented Generation (RAG), agentic AI, and document extraction pipelines. The system provides a rich Python SDK, a Typer-based CLI (`docling convert`), a service client for remote conversion via `docling-serve`, and an MCP server. It supports multiple output formats including Markdown, JSON, HTML, DocTags, and plain text. Architecturally, Docling employs a modular pipeline system with pluggable backends for each input format and a stage-based processing model. The core pipeline (`StandardPdfPipeline`, `VlmPipeline`, `AsrPipeline`) orchestrates page-level operations (preprocessing, OCR, layout detection, table structure recognition, picture classification, reading order) and document-level enrichments (code/formula extraction, heading hierarchy, picture description). Inference engines are abstracted behind factory patterns supporting ONNX Runtime, HuggingFace Transformers, vLLM, MLX, KServe v2, and remote APIs. The library integrates with over 20 OCR engines and vision-language models, and includes a plugin system for extensibility. Key entry points include the `DocumentConverter` class, the CLI (`docling.cli.main`), and the service client (`docling.service_client`). External dependencies span PDF libraries (pypdfium2, docling-parse), ML frameworks (torch, transformers, onnxruntime), OCR engines (Tesseract, EasyOCR, RapidOCR), and cloud services (KServe, Azure, S3). The codebase is actively maintained with frequent releases, extensive test coverage, and a clear migration path from legacy components.
GenieX is a production-grade, on-device generative AI inference SDK and CLI developed by Qualcomm for Snapdragon platforms. It provides a C core with a plugin architecture supporting two inference backends: llama.cpp (GGUF models on CPU/GPU/NPU) and QAIRT (pre-compiled bundles on Hexagon NPU). The system includes a Rust-based model manager that downloads, caches, and resolves models from Hugging Face, Qualcomm AI Hub, Docker Hub, and local filesystem, with resumable chunked downloads and progress reporting. It exposes multiple client interfaces: a Go CLI with subcommands for inference, an OpenAI-compatible HTTP server (streaming chat/completions, tool calls), Python bindings with high-level AutoModel factories, Android bindings via JNI/Kotlin, and Go bindings. The architecture is cleanly layered: C SDK → plugin backends → hardware, with a separate model manager process. The codebase demonstrates strong engineering practices: extensive test suites (unit, integration, device-level), CI/CD pipelines, versioned releases, and cross-platform support (Linux ARM64, Windows ARM64, Android). While primarily an application, it includes benchmark infrastructure and quality evaluation tools that could be used for research purposes.
Marker is a high-accuracy document-to-markdown converter that transforms PDFs, images, DOCX, PPTX, XLSX, HTML, and EPUB files into clean markdown, JSON, HTML, or chunked output. It uses a pipeline architecture: a Provider extracts raw text and layout from the input, Builders (layout, line, OCR) construct a structured document model, Processors apply domain-specific transformations (table reconstruction, equation extraction, footnote handling, LLM-based correction), and Renderers produce the final output format. The system is driven by the Surya VLM for layout analysis and OCR, with an optional LLM enhancement layer (Gemini, Claude, OpenAI, etc.) for complex tasks like table merging, inline math, and image descriptions. Key entry points include CLI tools (`marker`, `marker_single`), a FastAPI server (`marker_server`), a Streamlit GUI, and a Python API (`PdfConverter`, `TableConverter`, `OCRConverter`). External dependencies include surya-ocr, pdftext, pypdfium2, torch, transformers, and various LLM SDKs. The codebase is well-structured with a clear separation of concerns, extensive test coverage, and a benchmark suite (olmocr-bench) that validates quality scores (~76% overall balanced mode).
MinerU is a production-grade document parsing engine that converts PDFs, images, and Office documents (DOCX, PPTX, XLSX) into structured Markdown and JSON. It supports multiple backends: a fast pipeline (layout analysis + OCR + formula/table recognition), a high-accuracy hybrid mode combining VLM and OCR, and a pure VLM mode using vLLM/LMDeploy. The system is deployed as a CLI tool, FastAPI service, Gradio web UI, or Docker container, with integrations for RAG frameworks like LangChain, LlamaIndex, and Dify. The codebase is well-architected with clean abstractions for data I/O (file/S3), model initialization (singleton pattern), and output generation (client-side rendering). It handles 109 languages and includes specialized modules for mathematical formula recognition (UniMERNet), table structure recovery (SLANet+/UNet), and office document normalization. While the code is actively maintained and functionally rich, it carries some technical debt from its PaddleOCR heritage (vendored code, eval-based builders) and has a complex optional dependency matrix for hardware accelerators.
SGLang is a high-performance inference serving system for large language models (LLMs) and multimodal models (vision, video, audio, diffusion). It provides an OpenAI-compatible API, a Python frontend DSL for programmable inference, and supports a wide range of hardware backends (NVIDIA CUDA, AMD ROCm, Intel XPU, Ascend NPU, Apple MLX). The architecture is a distributed serving runtime (SRT) with a scheduler, model executor, custom CUDA/Triton kernels, and advanced features like speculative decoding (EAGLE, MTP, NGRAM), disaggregated prefill/decode, hierarchical KV cache (HiCache), and quantization (FP8, NVFP4, AWQ, GPTQ). It also includes a Rust-based model gateway for routing and a multimodal generation pipeline for diffusion models. The system is designed for production-scale deployment, powering trillions of tokens daily across hundreds of thousands of GPUs.
This repository provides a comprehensive, production-grade training framework for fine-tuning the Qwen-VL family of vision-language models (Qwen2-VL, Qwen2.5-VL, Qwen3-VL, Qwen3.5) using HuggingFace Transformers, Liger-Kernel, and DeepSpeed. It supports Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), Group Relative Policy Optimization (GRPO), and classification fine-tuning. The architecture is modular: dataset loaders handle mixed-modality (image/video/text) data with reasoning field support, custom trainer subclasses optimize memory and enable generation-based evaluation, and model wrappers provide sequence classification heads with configurable MLP bridges. Key engineering achievements include monkey-patched vision transformer forward passes for reduced memory usage, robust handling of DeepSpeed ZeRO-3 with dummy vision inputs for text-only batches, and a workaround for PEFT/DeepSpeed compatibility. The codebase is actively maintained, dependencies are pinned, and the project demonstrates strong craftsmanship with several reusable components.
This repository provides a minimal, from-scratch implementation of a Vision-Language Model (VLM) called MiniMind-V, designed for educational purposes and low-resource training. It ingests the ALLaVA-4V dataset (Parquet format with image bytes and JSON conversations) for pretraining (1.27M caption pairs) and supervised fine-tuning (2.9M mixed instruct/caption/text). The architecture follows a LLaVA-1.5 style: a frozen SigLIP2 visual encoder (256×256, 64 patch tokens) projects via a 2-layer MLP into a MiniMind LLM (with optional Mixture-of-Experts). Training freezes the visual encoder and trains only the projector and selected LLM layers, achieving ~3 RMB / 2 hours on a single RTX 3090. The codebase includes training scripts (pretrain and SFT), evaluation, a Gradio web demo, and model conversion to HuggingFace Transformers format. While primarily a research/educational project, it demonstrates a clean, modular VLM pipeline that can be adapted for other multimodal tasks.
Moondream is a lightweight, open-source vision-language model (VLM) library and inference framework. The core of the repository is a PyTorch model (`MoondreamModel`) that comprises a vision encoder, a text decoder (based on a transformer with optional MoE layers), and a region model for spatial grounding. It supports image captioning, visual question answering, object detection, pointing, and gaze detection through a unified API. The library is designed for efficient inference on both GPU and CPU, with support for quantization (via torchao) and LoRA-based fine-tuning. The codebase is cleanly structured into modular components for vision, text, and region processing, each with clear configuration dataclasses. The repository also includes a HuggingFace integration (`HfMoondream`) for easy loading via the transformers library, and several ready-to-use demos (Gradio apps, CLI samples). Beyond the core library, the repo contains a rich set of evaluation scripts (ChartQA, POPE, DocVQA, COCO mAP, etc.) that benchmark the model on standard VLM tasks, and finetuning examples that demonstrate how to adapt the model to custom datasets using reinforcement learning or supervised fine-tuning. Additionally, there are recipe applications (promptable video redaction, content moderation, gaze detection) that showcase practical use cases built on top of Moondream. While the primary character is an application/library, the presence of extensive evaluation and finetuning code gives it a hybrid research-engineering flavor. The code is well-maintained, with active commits and CI, and uses up-to-date dependencies, resulting in low technical drift.
Dolphin is a document image parsing tool that uses a two-stage analyze-then-parse approach with a VLM (Qwen2.5-VL). It takes images or PDFs, performs layout analysis to detect elements (text, tables, formulas, code, figures) with reading order, then parses each element in parallel batches. Outputs structured JSON and Markdown with visualized layout. The codebase is well-structured, offering CLI entry points for page-level, element-level, and layout-only parsing. It demonstrates good engineering practices with batching, coordinate mapping, and post-processing for formula/special characters. The core implementation wraps the Hugging Face Qwen2.5-VL model and includes utility modules for layout string parsing, PDF conversion, and markdown generation. The architecture is a CLI application with minimal coupling, making the utility modules potentially reusable across document parsing projects.
This repository is a large monorepo for a multimodal AI agent platform developed by ByteDance. It includes a desktop Electron application (UI-TARS Desktop) that allows users to control their computer using natural language via vision-language models (UI-TARS, GPT-4o, Claude, etc.). Additionally, it provides a CLI tool (Agent TARS) for running agents in browser/desktop environments, a server for multi-user agent sessions, a web UI for interacting with agents, MCP servers for browser/filesystem/commands, and a comprehensive SDK for building custom agents and operators. The architecture is heavily componentized with multiple packages for agent core, tool-calling engines (native, prompt-engineering, structured-outputs), browser automation (local, remote, sandbox), desktop automation (NutJS, ADB), and UI components. While the codebase demonstrates strong engineering with clean abstractions (event streams, plugin systems, strategy patterns), it also shows signs of rapid growth with some duplication (two agent server implementations, overlapping operator packages) and ongoing migrations. The repository is actively maintained with frequent releases and comprehensive test coverage across unit, integration, and e2e tests.
This repository is a comprehensive, curated list of anomaly detection (outlier detection) resources, maintained by the author of the PyOD library. It aggregates books, academic papers, tutorials, open-source toolkits, datasets, and relevant conferences/journals, covering tabular, time-series, graph, and text data modalities. The repository also includes two utility scripts: a downloader for paper PDFs and a robust URL checker for verifying links in the documentation. While the core value is the curated resource list itself, the utility scripts demonstrate clean engineering practices and are reusable for similar link-checking and batch download tasks. This is not a software application or data pipeline; it is a reference catalog for researchers and practitioners.
FluxVLA is a comprehensive, config-driven framework for training, evaluating, and deploying Vision-Language-Action (VLA) models for robotic manipulation. It supports multiple VLA architectures (OpenVLA, Pi0/Pi0.5, GR00T, Cosmos3, DreamZero, etc.) and robot platforms (LIBERO simulation, RoboCasa, Franka, UR3, ALOHA, Oli humanoid, Tron2). The codebase provides a modular pipeline with registries and builders for datasets, transforms, models (backbones, heads, projectors), optimizers, collators, and inference runners. Production capabilities include distributed training (FSDP/DDP), real-time ROS inference, custom CUDA/Triton kernels for performance, and evaluation with metrics reporting to Feishu. While primarily an engineering framework for applied robot learning, it also contains research contributions such as flow-matching action heads, reward models (ARM, SARM), and real-time chunking (RTC) guidance. The architecture is clean and extensible, though config proliferation and vendored third-party code introduce maintenance overhead.
The solo-cli is a command-line tool designed to deploy and manage Physical AI systems on local hardware. It integrates functionalities for serving large language models (via Ollama, vLLM, or llama.cpp), controlling robotic arms (through the LeRobot framework), authenticating with a custom model hub (Solo Hub), and fine-tuning models using external datasets. The application features a modular command structure with lazy-loaded dependencies to optimize startup time, cross-platform hardware detection for automated backend selection, and a layered configuration system. Additional utilities include Docker management, system information gathering, and filesystem-safe string handling for robotics data operations. The project also includes example implementations of Model Context Protocol (MCP) servers for various domains, demonstrating how to extend the CLI's capabilities for specialized AI workflows.
The repository implements LLaVA-VLA, a Vision-Language-Action model for robotic manipulation tasks. It extends the LLaVA framework to predict action sequences from visual inputs, proprioceptive state, and language instructions using a diffusion-based action head. The system includes training pipelines for fine-tuning on datasets like CALVIN, inference servers (Flask and Gradio) for real-time robot control, and evaluation tools for benchmarks. The architecture features modular vision encoders, projectors, and resamplers with factory patterns for flexibility, action tokenization for discrete action prediction, and integration with PEFT/DeepSpeed for efficient training.
Visual Spatial Tuning (VST) is a research framework for enhancing Vision-Language Models (VLMs) with human-like visuospatial abilities, from perception to reasoning. It implements a two-stage pipeline: supervised fine-tuning (SFT) on 4.1M spatial perception samples (VST-P) followed by reinforcement learning (RL) on 135K reasoning samples (VST-R). The framework extends to Vision-Language-Action (VLA) models for robotic manipulation evaluated on the LIBERO benchmark. The repository contains custom model implementations (Qwen2.5VL, Qwen3VL) with sequence parallelism and FSDP support, data preprocessing scripts, evaluation benchmarks (VSIBench, 3D object detection on SUN RGB-D, VLM benchmarks), and a distributed RL training system (verl) with reward functions and advantage estimation. While primarily a research codebase, it includes production-quality training infrastructure and reusable data utilities. The code is actively maintained with recent commits, but has medium technical drift due to tight version pinning and reliance on an internal ByteDance framework (VeOmni) that is not publicly available, which may hinder full reproducibility.
CameraBench is a benchmark and research project for evaluating camera motion understanding in videos, comparing classical Structure-from-Motion (SfM) methods and modern Vision-Language Models (VLMs). The repository serves as a landing page with download scripts for the test dataset (videos and labels from Hugging Face), a leaderboard showing model performance, and links to fine-tuned Qwen2.5-VL models. The actual evaluation and training code resides in a separate repository (t2v_metrics). The repo lacks a self-contained pipeline, dependency management, or data processing scripts; it is essentially a static site with download helpers. A notebook template for a ReLU activation function exercise is included but is unrelated to the core project. The benchmark is novel and published at NeurIPS 2025, but the codebase itself is minimal and not reusable beyond data retrieval.
This repository is a comprehensive educational resource focused on large language models (LLMs), reinforcement learning (RL), and related training algorithms. It contains over 100 original diagrams (SVG/PNG/PDF) covering topics such as LLM architecture, SFT, DPO, RLHF, PPO, GRPO, policy gradient, and reasoning optimization. The repo also includes a curated index of open-source LLMs and VLMs with links to papers, code, and configs. While not a traditional software application or data pipeline, it serves as a visual knowledge base for researchers and practitioners in AI alignment and model training. The only executable code is a few utility scripts for image processing; the main value lies in the meticulously crafted diagrams that explain complex concepts with clarity. The repo is actively maintained and frequently updated with new content.