MLog

A bilingual blog crafted for our own voice

All Posts

Filter by keyword, tag, and category.

LLM Training Optimization#Large Language Models#Memory Optimization#Orthogonal Transformation#Single-GPU Training#POET-X#ai-paper#paper-daily

POET-X: Memory-efficient Large Language Model Training by Scaling Orthogonal Transformation

Large language model training often faces memory bottlenecks. This paper proposes the POET-X framework, which significantly reduces computational overhead and memory footprint by scaling orthogonal equivalence transformations while maintaining training stability and generalization. Experiments show that POET-X can pre-train a billion-parameter LLM on a single Nvidia H100 GPU, whereas AdamW runs out of memory under the same conditions. This provides a highly valuable training solution for resource-constrained teams.

Mar 8, 20265 min
Artificial Intelligence and FinTech#AI#Multi-Agent#Quantitative Trading#LLM#Python#ai-auto#github-hot

Exploring the Application of AI Multi-Agents in Quantitative Trading: An Analysis of the ai-hedge-fund Project

virattt/ai-hedge-fund is a Python-based proof-of-concept AI hedge fund project. Through multi-agent collaboration, the system simulates the trading strategies of renowned investment masters, including Charlie Munger and Cathie Wood. Primarily intended for educational purposes, it aims to explore the potential of large language models in financial trading decisions. The project has currently garnered over 46,000 stars on GitHub.

Mar 7, 20266 min
Robot Learning#Robotics#Imitation Learning#Augmented Reality#Policy Iteration#Data Collection#ai-paper#paper-daily

RoboPocket: Instant Robot-Free Policy Iteration with Smartphones

Data collection efficiency in imitation learning has always been a bottleneck in robotics. This paper proposes the RoboPocket system, which utilizes ordinary smartphones and AR visual foresight technology to achieve instant robot-free policy iteration. By visualizing predicted trajectories through remote inference and combining asynchronous online fine-tuning, the system doubles data efficiency. It provides a novel, low-cost, and high-efficiency paradigm for large-scale robot data collection.

Mar 7, 20265 min
AI Development Tutorial#AI#MCP#LLM#Tutorial#Cross-language#Open-source#ai-auto#github-hot

Microsoft Open-Sources MCP for Beginners: A Practical Guide to Building Cross-Language AI Workflows

Microsoft's "MCP for Beginners" is an open-source course designed to help developers master the Model Context Protocol (MCP) through real-world, cross-language code examples in C#, Java, TypeScript, Rust, and Python. Focused on building modular, scalable, and secure AI workflows, the project has garnered over 14,000 stars on GitHub, making it an excellent starting point for AI developers entering the MCP ecosystem.

Mar 6, 20265 min
AI Development Tools#AI Coding Assistant#Multi-Agent#CLI Tool#Automated Development#TypeScript#ai-auto#github-hot

Codebuff: An Open-Source Terminal AI Coding Assistant Based on Multi-Agent Collaboration

Codebuff is an open-source terminal AI coding assistant that allows developers to modify codebases directly using natural language commands. Unlike tools relying on a single large model, it utilizes a multi-agent collaborative architecture (including file picker, planner, editor, and reviewer agents), outperforming Claude Code in official benchmarks. It aims to provide precise context understanding and code editing capabilities, ideal for terminal geeks needing automated refactoring and rapid development.

Mar 5, 20265 min
Artificial Intelligence#AI Agent#Claude#Research Automation#Large Language Models#Python#ai-auto#github-hot

GitHub 12K Star Hit: The Open-Source Skill Pack Turning Claude into an All-Around Scientist

With the explosion of AI Agent technology, K-Dense-AI's open-source project claude-scientific-skills has garnered over 12,000 stars on GitHub. This project provides Claude with an out-of-the-box scientific research and analysis skill pack, covering fields like bioinformatics, financial analysis, and materials science. It instantly transforms large language models into versatile AI scientists, significantly boosting the automation efficiency of research and engineering tasks.

Mar 5, 20263 min
Automated Publishing#AI#One-Click Publishing#Integration Testing#Automated Generation#Metadata

AI One-Click Publishing Integration Test Sample

This article is an integration test sample used to verify the AI one-click publishing feature. The main purpose is to test whether the system can smoothly and accurately automatically generate metadata such as English translations, content summaries, relevant tags, and article categories during the article publishing process. Through this sample, developers can confirm the stability and correctness of the automated workflow, ensuring the efficiency and quality of subsequent large-scale content publishing.

Mar 4, 20261 min
Artificial Intelligence#AI Agents#Multi-Agent System#Automated Workflow#Indie Hackers#Productivity Tools

Build Your Exclusive AI Digital Agency: Why Agency-Agents is Booming

Want to own an all-around AI agency? Agency-Agents provides you with various professional AI agents ranging from front-end development to community operations. Each agent has a unique personality, standardized workflows, and reliable delivery capabilities, allowing you to easily realize the entrepreneurial dream of a one-person army.

Mar 4, 20263 min