Awesome GitHub RepositoriesVision Model Inputs

Page Assist 是一个基于浏览器的 AI 集成工具，提供侧边栏界面，以便在浏览网页时与 AI 模型进行交互。它专注于隐私优先的聊天和网页内容分析，允许用户从活动网页中提取信息并进行查询，从而获得上下文感知的回复。 该项目通过本地 AI 集成脱颖而出，支持连接到本地托管的模型或私有 API 端点，无需依赖云服务即可处理数据。它还通过公共共享链接或自托管共享基础设施支持协作式 AI 对话，以实现更好的数据控制。 该系统包含浏览器自动化功能，例如通过将浏览器会话作为一组工具来控制标签页和与页面元素交互。它还通过网页截图和基于 DOM 的内容提取支持视觉处理，以进行全面的页面分析。

CodeCompanion is a Neovim plugin that brings large language model capabilities directly into the editor, enabling turn-based conversations with AI models in a dedicated chat buffer. It provides a comprehensive interface for interacting with LLMs, supporting multiple providers through a flexible adapter system that can route requests to various hosted or local language model services. The plugin distinguishes itself through its extensive context-sharing capabilities, allowing users to send buffer contents, visual selections, git diffs, LSP diagnostics, terminal output, quickfix lists, and view

Captum is an open-source library for explaining model predictions by attributing them to input features, neurons, and layers using gradient-based and perturbation-based methods. It provides a modular framework for implementing, evaluating, and combining a range of explanation techniques, including gradient-based attribution, perturbation-based analysis, game-theoretic Shapley value approximation, and surrogate model explanations, with support for parallelization and noise stabilization. The library distinguishes itself through its breadth of attribution methods and its support for advanced in

mmocr 是一个基于 PyTorch 的光学字符识别（OCR）框架，旨在训练和部署文本检测、识别和关键信息提取模型。它作为一个全面的场景文本检测和识别工具箱，提供用于定位文本区域并将视觉文本转换为机器编码字符串的专用库。 该项目的独特之处在于用于关键信息提取的研究框架和高级文本定位功能。这些包括使用 Transformer 的基于点的定位，以及使用参数化贝塞尔曲线来识别和转录任意形状的文本。 该框架涵盖了广泛的计算机视觉功能，包括用于增强和标准化多样化 OCR 数据集的流水线管理、具有分布式扩展的模型训练，以及使用标准 OCR 指标的性能评估。它还提供用于几何多边形操作和结果可视化的实用程序，以便根据真实标注审计预测。 该系统使用 Python 实现，并支持通过 Docker 环境打包进行安装。

This project is an AI API gateway and proxy that translates and normalizes requests between different AI model formats to ensure compatibility across client applications. It functions as a middleware service that can transform local command-line binaries into web services, allowing them to be triggered via HTTP requests. The system is distinguished by its ability to route multimodal text and image inputs and extract internal reasoning chains from model outputs to separate the chain of thought from the final answer. It includes an authentication manager that automatically cycles through multip

Tinker Cookbook is an open-source framework for fine-tuning large language models, supporting supervised learning, reinforcement learning, and parameter-efficient techniques like LoRA adapters. It provides a complete pipeline for aligning models with human preferences through multi-stage RLHF workflows, from supervised fine-tuning through preference optimization to reinforcement learning. The framework distinguishes itself through recipe-based training orchestration, where fine-tuning workflows are defined as composable recipe files that chain data loading, model configuration, and training l