Machine Learning Articles

Discover awesome-gpu-engineering, the curated resource hub transforming how developers master GPU engineering for AI acceleration. From CUDA kernels to thousand-GPU clusters, this comprehensive guide reveals the secrets of efficient AI systems.

Fine-tune LLMs, vision, audio & OCR models natively on Apple Silicon with mlx-tune's Unsloth-compatible API. Write once on Mac, deploy unchanged to CUDA clusters. Complete guide with real code examples.

LingBot-VLA is a pragmatic Vision-Language-Action foundation model with 20,000 hours of pre-training data, 2.8× faster training than alternatives, and state-of-the-art performance on GM-100 and RoboTwin benchmarks. Open weights, production-ready deployment.

Discover how ade-python from LandingAI revolutionizes document extraction with agentic AI workflows. Parse, split, and extract structured data from any document using Pydantic schemas, async processing, and intelligent document understanding.

Discover 20+ tools to automatically visualize neural network architectures from code. Stop drawing diagrams by hand and explore curated solutions for Keras, PyTorch, TensorFlow, LaTeX, and more.

PyMuPDF4LLM converts PDFs to LLM-ready Markdown in one line of Python. No GPU, no cloud costs, no tokens. Features smart OCR, layout-aware extraction, and native LangChain/LlamaIndex integrations at 250x lower cost than vision LLMs.

Discover FastFlowLM: run LLMs on AMD Ryzen AI NPUs with zero GPU required. 10x more power-efficient, 17MB runtime, 256k context windows, installs in 20 seconds. The Ollama alternative purpose-built for XDNA2.

CodeAct revolutionizes LLM agents by using executable Python code instead of JSON actions, achieving 20% higher success rates. This complete guide covers installation, real code examples from the official repository, and production deployment strategies.

Master LLMs from scratch with Hands-On Large Language Models — the official O'Reilly book repository with 300+ visuals, 12 executable chapters, and production-ready code for RAG, fine-tuning, and multimodal AI.

UltraShape-1.0 from PKU-YuanGroup revolutionizes 3D generation with a two-stage diffusion pipeline: coarse structure synthesis followed by voxel-based geometric refinement using RoPE spatial encoding. This complete guide covers installation, real code examples, and production deployment.

LiveBench is a contamination-free LLM benchmark with monthly question releases and objective scoring. Learn how to evaluate models honestly using this ICLR 2025 Spotlight Paper framework.

Discover WebShop, Princeton's groundbreaking NeurIPS 2022 environment for training AI agents to navigate e-commerce sites with 1.18M real products. Complete setup guide, code examples, and sim-to-real transfer insights included.

Open-source companion to O'Reilly's AI Systems Performance Engineering book. GPU optimization, distributed training, and inference scaling with PyTorch, CUDA, and Triton. 1,662 stars, Apache 2.0.

GibsonAI/memori is agent-native memory infrastructure that turns AI agent execution into structured, persistent state. LLM-agnostic with 15,590 GitHub stars, it integrates with existing data infrastructure and deploys across cloud, VPC, and on-premises.

TencentQQGYLab/AppAgent is an MIT-licensed Python framework that uses GPT-4V or Qwen-VL to automate Android apps through human-like tap and swipe interactions, learning from autonomous exploration or human demonstration.

Paper2Any is an open-source Python toolkit that converts research paper PDFs into editable figures, diagrams, and presentations. Built by OpenDCAI with 2,700+ GitHub stars, it supports 14 workflow modules including Paper2Figure, Paper2PPT, and Image2Drawio. Self-host via Docker or use standalone CLI scripts.

poloclub/transformer-explainer runs live GPT-2 in your browser with interactive visualization of every internal operation. Built by Georgia Tech researchers, MIT licensed, 8.2k stars.

mlcourse.ai is an open Machine Learning course by OpenDataScience, led by Yury Kashnitsky. It offers 10 weeks of self-paced learning from Pandas to gradient boosting, with Kaggle competitions, multilingual content, and optional bonus assignments.