The repository provides the official PyTorch implementation of DoRA (Weight-Decomposed Low-Rank Adaptation), a parameter-efficient fine-tuning method. It integrates with Hugging Face PEFT and includes training/evaluation scripts for vision-language and language tasks (e.g., VQA, ScienceQA, commonsense reasoning). The architecture is modular, with core components like DoRA tuners, adapter controllers, and low-rank layers. Entry points include training scripts (finetune.py, train_dora.py), evaluation scripts (evaluate.py, run_llava.py), and serving tools (gradio_web_server.py). External dependencies include PyTorch, Transformers, PEFT, bitsandbytes, and dataset-specific libraries.
This repository is a serverless web application that serves as a playground and API proxy for Google's Gemini Multimodal Live API. It provides a client-side interface for real-time text, audio, video, and screen-sharing interactions with the Gemini model. Crucially, it includes a backend proxy layer (deployable on Cloudflare Workers or Deno) that converts the native Gemini WebSocket/REST API into an OpenAI-compatible format, enabling integration with third-party clients like ChatBox or Cursor. The architecture is split between a static frontend (vanilla JS) and a lightweight serverless proxy that handles authentication, protocol translation, and streaming logic.
This repository is a comprehensive collection of examples, tutorials, and applications demonstrating how to use LanceDB (a vector database) for various AI/ML applications. It showcases patterns for retrieval-augmented generation (RAG), multimodal search, chatbots, recommendation systems, and AI agents. The repository contains both Python and JavaScript/TypeScript implementations, with many examples featuring web interfaces using Streamlit, Gradio, or React. While it serves as a valuable resource for learning LanceDB integration patterns, the quality varies across examples, and some content is marked as archived.
Pixel-Reasoner implements a two-stage training framework for Vision-Language Models (VLMs) to enable pixel-space reasoning. The system first performs instruction tuning on synthesized reasoning traces to familiarize models with visual operations (zoom-in, select-frame), then applies curiosity-driven reinforcement learning to balance exploration between pixel-space and textual reasoning. Targeting visual reasoning benchmarks (V* bench, TallyQA-Complex, InfographicsVQA, MVBench), it provides pre-trained models and evaluation scripts. The architecture combines adapted OpenRLHF components with custom VLM modules, Ray-based distributed training, and specialized data processors for multimodal input. Key entry points include training scripts (single/multi-node) and evaluation scripts, with data flowing from Hugging Face datasets through tokenization/image processing to PPO optimization loops.
This repository is a comprehensive Chinese-language educational tutorial and codebase for Retrieval-Augmented Generation (RAG) systems, created by Datawhale. It systematically covers the full RAG stack from fundamentals (data loading, chunking, embeddings) to advanced techniques (hybrid search, query construction, Text2SQL, reranking, agentic RAG, and Graph RAG). The code is organized by chapter, with runnable Python scripts demonstrating each concept using popular frameworks like LangChain, LlamaIndex, and LangGraph, alongside vector databases (Milvus, FAISS) and graph databases (Neo4j). While primarily an educational resource, the repository contains several well-crafted, reusable components such as a minimal Text2SQL agent framework, a clean LangGraph-based agentic RAG pattern, and a custom ColBERT reranker. The practical application is a Chinese recipe Q&A system ('尝尝咸淡'), which serves as a consistent domain for demonstrating various RAG techniques. The codebase is hybrid in nature, blending structured educational content with functional demo applications. The repository is actively maintained with a large community following. However, users should be aware of technical drift risks including vendored model code, hardcoded API dependencies, and the presence of AI-generated code sections. The core value lies in its structured curriculum and reusable RAG patterns rather than as a production-ready system.
This repository is a comprehensive family of GUI automation agents (Mobile-Agent series) developed by Tongyi Lab, Alibaba Group. It provides end-to-end solutions for automating interactions across mobile (Android), desktop (Windows/macOS/Linux), and web platforms using vision-language models. The system perceives GUI elements through screenshots, grounds natural language instructions to actionable commands (click, type, scroll), and executes tasks via platform-specific controllers. The architecture includes multiple agent versions (v1-v3.5, E, UI-S1, PC-Agent) with shared core modules for perception, action parsing, and controller interfaces, alongside training frameworks (VERL) and evaluation suites (Android World, MiniWoB, Mind2Web). Key entry points are version-specific run scripts (e.g., run_ma35.py) and API services for remote agent control. The system relies on external dependencies like ADB for Android, PyAutoGUI for desktop, Playwright for web, and VLMs (Qwen series) for reasoning.
MiniMind-O is a research codebase for training and evaluating a small-scale (0.1B parameter) omni-modal language model that processes text, audio, and image inputs and generates text or streaming speech. The architecture uses a Thinker–Talker dual-path design with frozen external encoders (SenseVoice, SigLIP2, Mimi codec, CAM++) and trainable projectors, plus a Multi-Token Prediction (MTP) head for audio codec tokens. The repository includes a complete training pipeline (SFT with DDP, mixed precision, checkpointing), a comprehensive evaluation CLI, a Gradio web demo, and a real-time Flask WebSocket demo with VAD and ASR for near-duplex interaction. Data is stored in parquet format with pre-encoded audio codes, images, and speaker embeddings. The project is primarily research-oriented, aiming to provide a transparent, reproducible baseline for omni-modal models, but also includes application-like components for interactive demos.
EchoMimicV3 is a unified multi-modal and multi-task human animation model that generates realistic human videos from audio and/or text prompts. Built upon the Wan2.1-Fun-V1.1-1.3B-InP base model, it supports both talking-head and full-body animation with two versions: preview (standard quality) and flash (optimized for speed and lower VRAM). The system provides Gradio-based interactive demos (app.py, app_mm.py) for real-time video generation from user-uploaded media, command-line inference scripts (infer_flash.py, infer_preview.py, run_flash.sh) for batch processing, and deployment options for ComfyUI, ModelScope, and HuggingFace. Core functionality centers on audio-driven lip-sync and facial animation, with optional text guidance for pose/scene control, leveraging diffusion models, transformer architectures, and multimodal encoders (CLIP for images, T5 for text, Wav2Vec2 for audio).
DeepSeek-OCR is an inference repository for a multimodal OCR model that combines vision encoders (SAM and CLIP) with a large language model (DeepSeek-V2/V3) to perform optical character recognition with layout awareness, supporting native and dynamic resolution processing. The system provides two inference paths: a Hugging Face Transformers integration for simple usage and a custom vLLM integration for high-throughput batched inference, including PDF processing and evaluation benchmarks. The architecture is a monolithic model definition with separate modules for vision encoding (deepencoder), image preprocessing (process), and vLLM glue code, all tightly coupled to the specific model architecture. The codebase is functional and demonstrates competent engineering for research-grade inference, but lacks production hardening, documentation, and consistent coding standards.
LLaVA-NeXT is a large multimodal model (LMM) research codebase that extends the LLaVA family to support single-image, multi-image, video, and interleaved tasks. It provides training pipelines for supervised fine-tuning (SFT), direct preference optimization (DPO), and reinforcement learning (PPO, GRPO) with custom vision encoders (CLIP, SigLIP, EVA-CLIP, MLCD) and modular projector/resampler components. The repository also includes a full RL training framework (EasyR1/verl) for large-scale GRPO training using Ray, FSDP, and vLLM, as well as evaluation scripts for video and interleaved benchmarks. While primarily a research codebase with extensive training configurations and experimental scripts, it also contains serving infrastructure (Gradio, SGLang, controller-worker) for deployment. The code is well-structured with reusable vision encoder wrappers, data protocols, and RL algorithm implementations, but suffers from hardcoded internal paths, multiple overlapping RL implementations, and legacy training pipelines that increase maintenance burden.
This repository contains a cross-platform mobile application (iOS and Android) that functions as an AI assistant for Ray-Ban Meta smart glasses. The app connects to the glasses via the Meta Wearables DAT SDK to access camera and microphone streams, then provides various AI-powered features including real-time conversations (Live AI), image recognition (Quick Vision), food nutrition analysis (LeanEat), real-time translation, and experimental RTMP streaming. The architecture follows MVVM patterns on both platforms, with SwiftUI/Jetpack Compose for UI, ViewModels for state management, and service layers handling API communications with multiple AI providers (Alibaba Cloud, Google Gemini, OpenRouter). Key features include secure API key storage, Bluetooth audio routing for glasses microphones, and integration with external systems like OpenClaw. The application is actively maintained with recent commits and demonstrates solid engineering practices, though the multi-platform nature and reliance on external AI services introduce some complexity.
Show-o and Show-o2 are research repositories for unified multimodal understanding and generation. They implement transformer-based models (1.3B, 1.5B, 7B parameters) that can perform text-to-image generation, image/video captioning, visual question answering, inpainting/extrapolation, and mixed-modality generation (interleaved text and images). The codebase is organized around two major model versions: Show-o uses discrete tokens with MAGVITv2 VAE and Phi-1.5 LLM, while Show-o2 moves to continuous latent space with WanVAE (3D causal VAE) and Qwen2.5 LLM. Training follows a multi-stage pipeline (pretraining, instruction tuning) with novel techniques like flow matching (via a modular `transport` package), dual-path vision encoders (SigLIP + patch embedding), and omni-attention masks that combine causal and bidirectional attention for multimodal sequences. The repository includes distributed training configs (DeepSpeed, multi-node), evaluation scripts for GenEval and DPG-Bench, and comprehensive YAML configuration files. This is primarily a data_research codebase: it produces research artifacts (models, papers at ICLR 2025 and NeurIPS 2025) rather than a production application. The code exhibits good modularity for a research project, with reusable components like the transport module, omni-attention masks, mixed dataloader, and attention masking utilities. However, it suffers from typical research-code issues: hardcoded internal infrastructure paths, reliance on non-public data pipelines (e.g., internal packages for RefinedWeb), and some code duplication between Show-o and Show-o2. Reproducibility is moderate—dependencies are pinned and configs are detailed, but many datasets are large-scale web sources that require manual setup, and the training scripts assume specific hardware (multi-node A100 clusters). The project is actively maintained (commits in early 2026) with a roadmap for future work, including mixed-modal generation and video support.
This repository is a curated collection of Jupyter notebooks and Python scripts that demonstrate fine-tuning, inference, quantization, and optimization techniques for state-of-the-art vision and multimodal models from the Hugging Face ecosystem. It covers a wide range of tasks including visual question answering, object detection, image classification, video understanding, multimodal retrieval, and preference alignment (DPO). The code is primarily educational, designed to be run interactively in Google Colab, with each notebook or script serving as a self-contained tutorial for a specific model or technique. Notable patterns include LoRA/QLoRA fine-tuning, knowledge distillation via a custom ImageDistilTrainer, any-to-any embedding for retrieval, and video frame downsampling for multimodal input. The repository is not a production library or application; it is a learning resource for engineers and researchers exploring vision-language models.
ExLlamaV2 is a high-performance inference library for running quantized large language models (LLMs) on consumer GPUs, supporting GPTQ and the custom EXL2 quantization formats. It provides a Python library with custom CUDA/C++ extensions for efficient attention, MLP, sampling, and KV cache management, along with advanced features like dynamic batching, paged attention, speculative decoding, LoRA, tensor parallelism, and multimodal (image/video) support. The library includes a WebSocket server for remote inference and a suite of example scripts demonstrating various use cases. While the codebase exhibits strong engineering with clean abstractions (e.g., module system, filter pattern, multi-backend attention) and highly optimized CUDA kernels, the repository has been archived in favor of ExLlamaV3, meaning no further updates or bug fixes are expected. This makes it a valuable reference for understanding quantized LLM inference internals but risky for new projects requiring ongoing support.
This repository is a personal cookbook and project showcase centered on the MiniCPM family of large language models (MiniCPM, MiniCPM-V, MiniCPM3). It contains over a dozen distinct application projects, including multi-modal OCR + visual grounding (OCR_VG), cross-modal image-text search, ReAct-style agent planning (agent_auto_plan), MBTI personality role-play, low-memory RAG (4G_memory_rag), AWQ quantization, embedding extraction, SFT/DPO fine-tuning pipelines, and Windows agent demos. Each project includes source code, shell scripts, and accompanying Chinese/English tutorials in markdown format. The codebase is primarily a collection of independent, loosely coupled modules rather than a unified application. It demonstrates practical usage of MiniCPM models for fine-tuning, deployment with vLLM/Ollama/llama.cpp, and integration with LangChain. The quality is functional but inconsistent: many modules contain hardcoded paths, duplicated logic, and lack tests, yet several components (ReAct data generation, image slicing utilities, LangChain wrapper) exhibit sufficient isolation and craftsmanship to be reusable across projects. The repo is actively maintained with a recent commit, but dependency on unmerged forks and version-specific model files poses a medium technical drift risk.
The repository is a simple Flask web application demonstrating interaction with the Google Gemini API for text and multimodal chat. It provides a UI for sending text messages and uploading images, maintaining a chat session via the Gemini API and streaming responses using Server-Sent Events. The architecture is monolithic with backend logic in app.py, served frontend static files, and a single HTML template. Key endpoints include '/' for the main page, '/upload' for file handling, '/chat' for message reception, and '/stream' for response streaming. While functional for demonstration, the application relies on global variables for state management (chat session, pending message, pending image), making it unsuitable for production multi-threaded environments without modification. The code is straightforward and serves as an educational starting point for Gemini API integration, but lacks reusable components due to tight coupling and simplicity-focused design. Dependencies are current, and the repository shows recent maintenance activity.
This repository contains the research code for Video-SALMONN-2, a family of audio-visual large language models for video understanding, developed by Tsinghua University and ByteDance. It includes three generations: Video-SALMONN-2 (based on LLaVA), Video-SALMONN-2+ (based on Qwen2.5-VL), and Video-SALMONN-2 Pro (based on Qwen3-VL). The core contribution is a caption-enhanced training paradigm that tightly aligns audio and visual signals. The codebase provides full training pipelines for supervised fine-tuning (SFT) and direct preference optimization (DPO), along with evaluation scripts, a demo mode, and a vLLM-based serving plugin for deployment. While the project is predominantly research-focused, it includes application-like components such as CLI inference and OpenAI-compatible serving. The code exhibits high duplication across the three model variants, hardcoded paths to internal cluster storage, and brittle dependency pinning, making it challenging to reproduce outside the original environment. However, several components—such as the multimodal batch sampler, audio-video processing utilities, and QFormer-based fusion—are well-structured and potentially reusable for other multimodal research projects.