仓库/Skill 库

100+ LLM 与 VLM 的统一高效微调（ACL 2024）Unified Efficient Fine-Tuning of 100+ LLMs & VLMs (ACL 2024)

基于 PyTorch 的 Ultralytics YOLOv5，用于目标检测、实例分割、分类、训练与导出。Ultralytics YOLOv5 in PyTorch for object detection, instance segmentation, classification, training, and export.

CLIP（对比语言-图像预训练），根据图像预测最相关的文本片段CLIP (Contrastive Language-Image Pretraining), Predict the most relevant text snippet given an image

FinceptTerminal 是一款现代化金融应用，提供高级市场分析、投资研究与经济数据工具，旨在以友好交互环境支持探索式分析与数据驱动决策。FinceptTerminal is a modern finance application offering advanced market analytics, investment research, and economic data tools, designed for interactive exploration and data-driven decision-making in a user-friendly environment.

持续进行的大规模 Transformer 模型训练研究。Ongoing research training transformer models at scale

20+ 高性能 LLM，并附带大规模预训练、微调与部署的 recipes20+ high-performance LLMs with recipes to pretrain, finetune and deploy at scale.

训练你自己的 LLM 的简洁方法，覆盖从数据下载到文本生成A straightforward method for training your LLM, from downloading data to generating text.

ClearML —— 以自动化 CI/CD 简化 AI 工作负载。一站式 MLOps/LLMOps 解决方案，涵盖实验管理、数据管理、pipeline、编排、调度与服务。ClearML - Auto-Magical CI/CD to streamline your AI workload. Experiment Management, Data Management, Pipeline, Orchestration, Scheduling & Serving in one MLOps/LLMOps solution

Chronos：面向时间序列预测的预训练模型Chronos: Pretrained Models for Time Series Forecasting

在 Kubernetes 上借助现代加速器实现 SOTA 推理性能。Achieve state of the art inference performance with modern accelerators on Kubernetes

可扩展至上千个微调 LLM 的 Multi-LoRA 推理服务器Multi-LoRA inference server that scales to 1000s of fine-tuned LLMs

厂商无关的训练、推理与 Agent 负载编排，覆盖 NVIDIA、AMD、TPU 和 Tenstorrent，部署在云端、Kubernetes 和裸金属环境上Vendor-agnostic orchestration for training, inference and agentic workloads across NVIDIA, AMD, TPU, and Tenstorrent on clouds, Kubernetes, and bare metal.

轻松微调、构建并部署开源 LLM！🏗️ Fine-tune, build, and deploy open-source LLMs easily!

一款小型 macOS 应用，将 Caps Lock 键变为物理防休眠开关，用于合盖状态下的 MacBook 工作。A tiny macOS app that turns Caps Lock into a physical keep-awake switch for closed-lid MacBook work.

Windows 托盘上的 Kimi Code 用量监控：左键看额度，低额变红提醒。Tauri 2 + Rust，安装包仅 4 MB。

保存文章、AI 研究来源和 newsletter 邮件链接。准备好时再阅读。无第三方追踪。由 js-cookie 的作者基于其 10 年个人阅读系统打造。Save articles, AI research sources, and newsletter email links. Read them when you're ready. No third-party tracking. Built by the creator of js-cookie from a 10-year personal reading system.

Promovolve 是一项试图找回杂志广告昔日魅力的尝试。Promovolve is an attempt to get back what magazine advertising had!

🤖 利用 TRPO、PPO、DPO、GRPO、DAPO、GSPO 等先进算法提升强化学习的稳定性与效率，优化策略训练。🤖 Enhance reinforcement learning stability and efficiency with advanced algorithms like TRPO, PPO, DPO, GRPO, DAPO, and GSPO for optimized policy training.

MedTrack 是基于 React 与 Spring Boot 构建的全栈医疗设备管理系统，通过基于角色的安全访问简化库存、维护与设备订购流程。MedTrack is a full-stack medical equipment management system built with React and Spring Boot. It streamlines inventory, maintenance, and equipment orders through secure role-based access.

🐙 通过 Python 与 PyTorch 实战代码深入理解 LLM 内部机制，掌握 LLM 架构、训练及核心原理。🐙 Explore the inner workings of large language models with hands-on Python and PyTorch code. Master LLM architecture, training, and mechanisms for AI systems.

准确捕获 macOS 听写内容，不改写你的措辞，保留真实输入并规避常见应用问题Capture macOS dictation accurately without rewriting your words, keeping your input true to what you speak and avoiding common app issues.

🖥️ 通过终端中键盘优先的 Kanban board 简化工作流，实现快速、专注的任务管理。🖥️ Streamline your workflow with a keyboard-first Kanban board in your terminal for fast, focused task management.

Cypress AI 演示Cypress AI Demo

用于微调的合成数据样本集合，旨在为更大规模合成数据集提供种子数据This is a collection of synthetically generated datapoints for fine tuning meant for seeding larger synthetic datasets.

软件工程作品集 | Web 开发 | 桌面应用开发Software Engineering Portfolio | Web Development | Desktop App Development

Embrace the choas（拥抱混乱）Embrace the choas

AUA AI Hub：为安提瓜美国大学医学院策划的 AI 参考与新闻站点。AUA AI Hub: curated AI reference and news site for the American University of Antigua College of Medicine

语言驱动、无锁的分布式工作流引擎 — 用 FFL 编写工作流，运行时负责执行、依赖解析、恢复和热部署。Language-directed, lock-free distributed workflow engine — write workflows in FFL, the runtime handles execution, dependency resolution, recovery, and live deployment.

探索高达 400B 参数的先进稀疏 Mixture-of-Experts 模型，采用创新训练技术实现卓越性能🚀 Explore advanced sparse Mixture-of-Experts models with up to 400B parameters, featuring innovative training techniques for superior performance.

软件开发software development

借助这款自托管、本地优先的应用，跨多个账户与多种货币追踪收入和支出。Track income and expenses across multiple accounts and currencies with this self-hosted, local-first application.

用前瞻性奖励训练面向动机式访谈的小型 LLM 治疗师：Preference Tree Optimization (PTO) vs GRPO，配合模拟患者和 LLM 评委。里赫曼大学硕士论文。Training small LLM therapists for Motivational Interviewing with look-ahead rewards: Preference Tree Optimization (PTO) vs GRPO, with simulated patients and LLM judges. Master's thesis, Reichman University.

这个仓库展示了我的技能、项目，以及从数据工程转型为数据科学家角色的持续学习历程。This repository showcases my skills, projects, and continuous learning journey as I transition from Data Engineering to a Data Scientist role.

从零开始在 PyTorch 中构建 decoder-only Transformer，涵盖分词、预训练、监督微调，以及基于核心张量操作的对齐。Build decoder-only Transformers from scratch in PyTorch, covering tokenization, pretraining, supervised fine-tuning, and alignment using core tensor operations.

OptiBuild 工程案例研究——一款 AI 驱动的施工现场管理平台：工程量清单解析、成本核算、进度报告与按角色划分的数据权限。Engineering case study of OptiBuild, an AI-enabled construction-site management platform: quantity-survey parsing, costing, work-progress statements and role-scoped data.

AI 工程师与全栈开发者——从架构到生产，独立设计并交付完整的 AI-native 与全栈系统。AI Engineer & Full-Stack Developer — I design and ship complete AI-native and full-stack systems, from architecture to production.