{"id":107075,"url":"https://github.com/umitkacar/awesome-tinyml","name":"awesome-tinyml","description":"TinyML \u0026 Edge AI: On-device inference, model quantization, embedded ML, ultra-low-power AI for microcontrollers and IoT devices.","projects_count":227,"last_synced_at":"2026-08-07T11:00:22.158Z","repository":{"id":158690224,"uuid":"336525398","full_name":"umitkacar/awesome-tinyml","owner":"umitkacar","description":"TinyML \u0026 Edge AI: On-device inference, model quantization, embedded ML, ultra-low-power AI for microcontrollers and IoT devices.","archived":false,"fork":false,"pushed_at":"2025-11-10T09:48:13.000Z","size":370,"stargazers_count":52,"open_issues_count":0,"forks_count":4,"subscribers_count":4,"default_branch":"main","last_synced_at":"2026-07-19T04:05:42.269Z","etag":null,"topics":["arduino","cortex-m","edge-ai","edge-computing","embedded-ml","embedded-systems","esp32","iot","knowledge-distillation","microcontroller","model-optimization","neural-network-compression","on-device-inference","pruning","quantization","raspberry-pi","tensorflow-lite","tiny-deep-learning","tinyml","ultra-low-power"],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/umitkacar.png","metadata":{"files":{"readme":"README.md","changelog":"CHANGELOG.md","contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2021-02-06T11:49:23.000Z","updated_at":"2026-07-09T07:44:03.000Z","dependencies_parsed_at":"2024-01-03T13:28:52.302Z","dependency_job_id":"3d4dc6be-2fb9-473a-ac23-be501eede383","html_url":"https://github.com/umitkacar/awesome-tinyml","commit_stats":null,"previous_names":["umitkacar/awesome-ai-edge-computing-tiny-embedded","umitkacar/awesome-tinyml"],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/umitkacar/awesome-tinyml","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/umitkacar%2Fawesome-tinyml","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/umitkacar%2Fawesome-tinyml/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/umitkacar%2Fawesome-tinyml/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/umitkacar%2Fawesome-tinyml/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/umitkacar","download_url":"https://codeload.github.com/umitkacar/awesome-tinyml/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/umitkacar%2Fawesome-tinyml/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":36369422,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-08-06T04:43:03.162Z","status":"online","status_checked_at":"2026-08-07T02:00:06.708Z","response_time":57,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"created_at":"2026-01-16T15:23:26.829Z","updated_at":"2026-08-07T11:00:22.158Z","primary_language":null,"list_of_lists":false,"displayable":true,"categories":["🌐 **Edge Deployment Frameworks**","⚡ **State Space Models - Efficient Transformers**","🚀 **Quick Start \u0026 Development**","📚 **Research Papers \u0026 Academic Resources**","🔧 **Model Compression \u0026 Optimization**","🎯 **TinyML \u0026 MCU-specific Advances**","⚙️ **Compilers \u0026 Low-Level Frameworks**","🚀 **Inference Frameworks \u0026 Runtimes**","🛠️ **Implementation Resources \u0026 Tools**","🔧 **Utility Frameworks \u0026 Tools**","🤖 **Small Language Models (SLMs) for Edge**","🔥 **SOTA Models \u0026 Algorithms (2024-2025)**","🔩 **Hardware Acceleration \u0026 Platforms**","💼 **Industry \u0026 Commercial Solutions**","🖼️ **Additional Model Architectures**","🧠 **Computing Architectures \u0026 APIs**","🎓 **Contributing \u0026 Community**"],"sub_categories":["🍎 **CoreML** - Apple","🐍 **Mamba**","🛠️ **Development Setup**","📖 **Foundational Surveys (2024-2025)**","📉 **Advanced Quantization Techniques**","📱 **NCNN** - Tencent","🧠 **MCUNet Series** - MIT HAN Lab","⚡ **XNNPack** - Google","⚡ **TensorRT-LLM**","🔨 **LLVM**","🎯 **YOLO Implementations**","🔷 **ARM-NN**","🚀 **FastDeploy** - PaddlePaddle","🔧 **TVM** - Apache","🎬 **VQRF** - Video Compression","💎 **DeepSparse \u0026 SparseML** - Neural Magic","🧠 **CMSIS-NN**","🔧 **MACE** - Xiaomi","🦙 **Meta Llama 3.2**","🔷 **ONNX Runtime**","🧠 **Microsoft Phi-3**","📦 **Installation**","🎓 **Community**","🎯 **Features**","🎯 **Object Detection Models**","📱 **Efficient Vision Models for Edge**","🦙 **TinyLlama**","🌟 **Google Gemini Nano**","📷 **MobileVLM**","📱 **eMamba**","📄 **vLLM**","🦙 **ExecuTorch**","💻 **llama.cpp**","🔬 **Neural Architecture Search (NAS)**","🎓 **Knowledge Distillation \u0026 Pruning**","🔬 **TinyDL (Tiny Deep Learning)**","🖥️ **Edge AI Platforms**","📱 **Mobile Deployment Targets**","📉 **ONNX Runtime Quantization**","⚡ **TensorRT**","📱 **Samsung ONE**","🚀 **Deeplite**","👁️ **OpenCV**","🎯 **PP-PicoDet**","🔬 **EtinyNet**","📊 **Repository Stats**"],"readme":"\u003cdiv align=\"center\"\u003e\n\n# 🚀 AI Edge Computing \u0026 TinyML\n### *Comprehensive Guide to State-of-the-Art Edge AI*\n\n\u003cimg src=\"https://readme-typing-svg.demolab.com?font=Fira+Code\u0026size=32\u0026duration=2800\u0026pause=1000\u0026color=00D9FF\u0026center=true\u0026vCenter=true\u0026width=940\u0026lines=AI+Edge+Computing+%26+TinyML;Ultra-Low+Power+AI+Systems;Deploy+AI+on+Embedded+Devices;Real-Time+Inference+at+the+Edge\" alt=\"Typing SVG\" /\u003e\n\n[![GitHub stars](https://img.shields.io/github/stars/umitkacar/ai-edge-computing-tiny-embedded?style=for-the-badge\u0026logo=github\u0026color=yellow)](https://github.com/umitkacar/ai-edge-computing-tiny-embedded/stargazers)\n[![GitHub forks](https://img.shields.io/github/forks/umitkacar/ai-edge-computing-tiny-embedded?style=for-the-badge\u0026logo=github\u0026color=blue)](https://github.com/umitkacar/ai-edge-computing-tiny-embedded/network/members)\n[![License](https://img.shields.io/badge/License-MIT-green.svg?style=for-the-badge\u0026logo=opensourceinitiative)](LICENSE)\n[![Latest Update](https://img.shields.io/badge/Updated-January_2025-ff69b4?style=for-the-badge\u0026logo=clockify)](https://github.com/umitkacar/ai-edge-computing-tiny-embedded)\n\n---\n\n### 🌟 **Latest Update: January 2025**\n\u003e **Production-Ready Python Implementation** with modern tooling (Hatch, Ruff, Mypy)\n\u003e **62/62 Tests Passing** • **81.76% Coverage** • **Zero Security Issues**\n\u003e **State-of-the-Art Algorithms \u0026 Trends** for Edge AI and Embedded Systems\n\n\u003c/div\u003e\n\n---\n\n## 📋 **Table of Contents**\n\n\u003ctable\u003e\n\u003ctr\u003e\n\u003ctd width=\"33%\" valign=\"top\"\u003e\n\n### 🚀 **Getting Started**\n- [📦 Installation](#-installation)\n- [🛠️ Development Setup](#%EF%B8%8F-development-setup)\n- [📊 Project Structure](#-project-structure)\n- [✅ Quality Assurance](#-quality-assurance)\n- [🎯 Features \u0026 Examples](#-features)\n\n\u003c/td\u003e\n\u003ctd width=\"33%\" valign=\"top\"\u003e\n\n### 🔥 **Core Topics**\n- [🎯 SOTA Models 2024-2025](#-sota-models--algorithms-2024-2025)\n- [👁️ Object Detection](#-object-detection-models)\n- [🤖 Small Language Models](#-small-language-models-slms-for-edge)\n- [⚡ State Space Models](#-state-space-models---efficient-transformers)\n\n\u003c/td\u003e\n\u003ctd width=\"33%\" valign=\"top\"\u003e\n\n### 🛠️ **Frameworks \u0026 Tools**\n- [🚀 Inference Frameworks](#-inference-frameworks--runtimes)\n- [🔧 Model Optimization](#-model-compression--optimization)\n- [💻 Hardware Platforms](#-hardware-acceleration--platforms)\n- [🌐 Deployment Tools](#-edge-deployment-frameworks)\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"33%\" valign=\"top\"\u003e\n\n### 📚 **Documentation**\n- [📄 CHANGELOG.md](CHANGELOG.md)\n- [📚 LESSONS-LEARNED.md](LESSONS-LEARNED.md)\n- [🔧 DEVELOPMENT.md](DEVELOPMENT.md)\n\n\u003c/td\u003e\n\u003ctd width=\"33%\" valign=\"top\"\u003e\n\n### 📚 **Resources**\n- [🎯 TinyML \u0026 MCU](#-tinyml--mcu-specific-advances)\n- [⚙️ Compilers](#%EF%B8%8F-compilers--low-level-frameworks)\n- [📄 Research Papers](#-research-papers--academic-resources)\n\n\u003c/td\u003e\n\u003ctd width=\"33%\" valign=\"top\"\u003e\n\n### 🎓 **Community**\n- [🤝 Contributing](#-contributing--community)\n- [📊 Repository Stats](#-repository-stats)\n- [🏷️ Keywords](#%EF%B8%8F-keywords)\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/table\u003e\n\n---\n\n\u003cdiv align=\"center\"\u003e\n\n## 🚀 **Quick Start \u0026 Development**\n\n![Python](https://img.shields.io/badge/Python-3.11+-3776AB?style=for-the-badge\u0026logo=python\u0026logoColor=white)\n![Hatch](https://img.shields.io/badge/Hatch-Build_System-4051B5?style=for-the-badge\u0026logo=pypi\u0026logoColor=white)\n![Tests](https://img.shields.io/badge/Tests-62%2F62_Passing-success?style=for-the-badge\u0026logo=pytest\u0026logoColor=white)\n![Coverage](https://img.shields.io/badge/Coverage-81.76%25-brightgreen?style=for-the-badge\u0026logo=codecov\u0026logoColor=white)\n![Type](https://img.shields.io/badge/Type_Checked-Mypy_Strict-blue?style=for-the-badge\u0026logo=python\u0026logoColor=white)\n\n\u003c/div\u003e\n\n### 📦 **Installation**\n\nThis project uses modern Python tooling with [Hatch](https://hatch.pypa.io/) for dependency management and development workflows.\n\n```bash\n# Clone the repository\ngit clone https://github.com/umitkacar/ai-edge-computing-tiny-embedded.git\ncd ai-edge-computing-tiny-embedded\n\n# Install dependencies (using hatch)\npip install hatch\n\n# Run tests\nhatch run test\n\n# Run full CI pipeline\nhatch run ci\n```\n\n### 🛠️ **Development Setup**\n\n**Modern Python Stack:**\n- **Build System:** [Hatch](https://hatch.pypa.io/) - Modern Python project manager\n- **Linting:** [Ruff](https://docs.astral.sh/ruff/) - Ultra-fast Python linter (100x faster than flake8)\n- **Formatting:** [Black](https://black.readthedocs.io/) - The uncompromising code formatter\n- **Type Checking:** [Mypy](https://mypy.readthedocs.io/) - Static type checker (strict mode)\n- **Testing:** [Pytest](https://docs.pytest.org/) - Comprehensive test framework\n- **Security:** [Bandit](https://bandit.readthedocs.io/) - Security vulnerability scanner\n- **Pre-commit:** Automated quality checks on commit/push\n\n**Available Commands:**\n\n```bash\n# Linting \u0026 Formatting\nhatch run lint          # Run Ruff linter\nhatch run format        # Format code with Black\nhatch run format-check  # Check formatting without changes\n\n# Type Checking\nhatch run type-check    # Run Mypy strict type checking\n\n# Testing\nhatch run test                    # Run tests (sequential)\nhatch run test-parallel           # Run tests with auto workers\nhatch run test-parallel-cov       # Parallel tests with coverage\n\n# Security\nhatch run security      # Run Bandit security audit\n\n# Complete CI Pipeline\nhatch run ci           # Run all checks (format, lint, type-check, security, test)\n```\n\n### 📊 **Project Structure**\n\n```\nai-edge-computing-tiny-embedded/\n├── src/ai_edge_tinyml/          # Source code (src layout)\n│   ├── __init__.py              # Package initialization\n│   ├── quantization.py          # INT8/INT4/FP16 quantization\n│   ├── model_optimizer.py       # Model optimization pipeline\n│   ├── utils.py                 # Utility functions\n│   └── py.typed                 # PEP 561 marker (typed package)\n├── tests/                       # Test suite (62 tests, 81.76% coverage)\n│   ├── conftest.py              # Pytest configuration \u0026 fixtures\n│   ├── test_quantization.py     # Quantization tests (21 tests)\n│   ├── test_model_optimizer.py  # Optimizer tests (19 tests)\n│   └── test_utils.py            # Utility tests (22 tests)\n├── pyproject.toml               # Project configuration (single source of truth)\n├── .pre-commit-config.yaml      # Pre-commit hooks configuration\n├── CHANGELOG.md                 # Detailed change history\n├── LESSONS-LEARNED.md           # Best practices \u0026 insights\n├── DEVELOPMENT.md               # Development guidelines\n└── README.md                    # This file\n```\n\n### ✅ **Quality Assurance**\n\nThis project maintains production-ready code quality:\n\n| Check | Status | Details |\n|-------|--------|---------|\n| **Ruff Linting** | ✅ PASS | 50+ rules, zero errors |\n| **Black Formatting** | ✅ PASS | Line length: 100 |\n| **Mypy Type Check** | ✅ PASS | Strict mode enabled |\n| **Bandit Security** | ✅ PASS | 0 vulnerabilities |\n| **Test Suite** | ✅ PASS | 62/62 tests passing |\n| **Code Coverage** | ✅ PASS | 81.76% (exceeds 80%) |\n| **Pre-commit Hooks** | ✅ PASS | 15+ automated checks |\n\n**Test Results:**\n```\ntests/test_quantization.py      21 passed\ntests/test_model_optimizer.py   19 passed\ntests/test_utils.py             22 passed\n━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\nTotal: 62 passed in 0.50s ✅\nCoverage: 81.76% (exceeds 80% threshold) ✅\n```\n\n### 🔒 **Security**\n\n- **Bandit Security Audit:** Zero vulnerabilities detected\n- **Type Safety:** Full type annotations with mypy strict mode\n- **Dependency Scanning:** Automated security checks in CI\n- **Pre-commit Hooks:** Security validations before commit\n\n### 📚 **Documentation**\n\n- **[CHANGELOG.md](CHANGELOG.md)** - Detailed version history and changes\n- **[LESSONS-LEARNED.md](LESSONS-LEARNED.md)** - Best practices, insights, and technical decisions\n- **[DEVELOPMENT.md](DEVELOPMENT.md)** - Comprehensive development guidelines\n- **API Documentation:** Auto-generated from Google-style docstrings\n\n### 🎯 **Features**\n\n**Quantization Support:**\n- ✅ INT8 Quantization (8-bit integers)\n- ✅ INT4 Quantization (4-bit integers)\n- ✅ FP16 Quantization (16-bit floats)\n- ✅ Dynamic Quantization\n- ✅ Symmetric \u0026 Asymmetric modes\n- ✅ Per-tensor \u0026 per-channel quantization\n\n**Model Optimization:**\n- ✅ Weight quantization with 6 different modes\n- ✅ Compression ratio analysis\n- ✅ Model size calculation\n- ✅ Type-safe APIs with full annotations\n- ✅ Comprehensive error handling\n\n**Example Usage:**\n\n```python\nimport numpy as np\nfrom ai_edge_tinyml import Quantizer, QuantizationConfig, QuantizationMode\n\n# Create quantization config\nconfig = QuantizationConfig(\n    mode=QuantizationMode.INT8,\n    symmetric=True,\n    per_channel=False\n)\n\n# Initialize quantizer\nquantizer = Quantizer(config)\n\n# Quantize weights\nweights = np.random.randn(100, 100).astype(np.float32)\nquantized = quantizer.quantize(weights)\n\n# Dequantize for inference\ndequantized = quantizer.dequantize(quantized)\n\n# Calculate compression\nfrom ai_edge_tinyml.utils import calculate_compression_ratio\nratio = calculate_compression_ratio(weights, quantized)\nprint(f\"Compression ratio: {ratio:.2f}x\")\n```\n\n---\n\n\u003cdiv align=\"center\"\u003e\n\n## 🔥 **SOTA Models \u0026 Algorithms (2024-2025)**\n\n![AI Edge](https://img.shields.io/badge/AI-Edge_Computing-00D9FF?style=for-the-badge\u0026logo=tensorflow\u0026logoColor=white)\n![TinyML](https://img.shields.io/badge/TinyML-Embedded_AI-FF6B6B?style=for-the-badge\u0026logo=arduino\u0026logoColor=white)\n![SOTA](https://img.shields.io/badge/SOTA-2024--2025-4ECDC4?style=for-the-badge\u0026logo=artifacthub\u0026logoColor=white)\n\n\u003c/div\u003e\n\n---\n\n### 🎯 **Object Detection Models**\n\n\u003ctable\u003e\n\u003ctr\u003e\n\u003ctd width=\"50%\"\u003e\n\n#### 🥇 **YOLOv11 (YOLO11)**\n![Release](https://img.shields.io/badge/Release-November_2024-brightgreen?style=flat-square\u0026logo=github)\n![Status](https://img.shields.io/badge/Status-SOTA-gold?style=flat-square\u0026logo=hackthebox)\n\n\u003e 🚀 State-of-the-art real-time object detection with transformer-based improvements\n\n**✨ Key Features:**\n- ⚡ Transformer-based backbone with C3k2 blocks\n- 🎯 Partial Self-Attention (PSA) mechanism\n- 🔥 NMS-free training with dual label assignment\n- 📉 **25-40% lower latency** vs YOLOv10\n- 📊 **10-15% improvement** in mAP\n- ⚡ **60+ FPS** processing capability\n\n**📚 Resources:**\n```bash\n📖 Ultralytics Docs → https://docs.ultralytics.com/models/\n📄 YOLO Evolution → https://arxiv.org/html/2510.09653v2\n```\n\n\u003c/td\u003e\n\u003ctd width=\"50%\"\u003e\n\n#### 🥈 **YOLOv10**\n![Release](https://img.shields.io/badge/Release-May_2024-blue?style=flat-square\u0026logo=github)\n![NMS](https://img.shields.io/badge/NMS-Free-orange?style=flat-square\u0026logo=lightning)\n\n\u003e ⚡ Eliminates NMS for end-to-end real-time detection\n\n**📊 Performance Metrics:**\n- 🔸 **YOLOv10s**: 1.8x faster than RT-DETR-R18\n- 🔸 **YOLOv10b**: 46% less latency, 25% fewer parameters\n- 🔸 **mAP Range**: 38.5 - 54.4\n\n**📚 Resources:**\n```bash\n📄 Paper → https://arxiv.org/pdf/2405.14458\n📖 Docs → https://docs.ultralytics.com/models/yolov10/\n```\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/table\u003e\n\n---\n\n#### 🤖 **RT-DETR \u0026 RT-DETRv2**\n![Transformer](https://img.shields.io/badge/Architecture-Transformer-blueviolet?style=for-the-badge\u0026logo=pytorch)\n![Real-Time](https://img.shields.io/badge/Real--Time-Detection-success?style=for-the-badge\u0026logo=speedtest)\n\n\u003e 🎯 First practical real-time detection transformer\n\n| Model | AP Score | FPS | Device |\n|-------|----------|-----|--------|\n| RT-DETR | **53.1%** | 108 | NVIDIA T4 |\n| RT-DETRv2 | **\u003e55%** | 108+ | NVIDIA T4 |\n\n**🔗 Resources:**\n- 📊 [RT-DETR vs YOLO11 Comparison](https://docs.ultralytics.com/compare/rtdetr-vs-yolo11/)\n\n---\n\n### 📱 **Efficient Vision Models for Edge**\n\n\u003cdiv align=\"center\"\u003e\n\n```mermaid\ngraph LR\n    A[🖼️ Input Image] --\u003e B[📱 MobileNetV4]\n    A --\u003e C[⚡ EfficientViT]\n    B --\u003e D[🎯 87% Accuracy]\n    C --\u003e E[🔥 3.8ms Latency]\n    D --\u003e F[📲 Edge TPU]\n    E --\u003e F\n    style A fill:#e1f5ff\n    style B fill:#ffe1f5\n    style C fill:#f5ffe1\n    style D fill:#ffe1e1\n    style E fill:#e1ffe1\n    style F fill:#ffd700\n```\n\n\u003c/div\u003e\n\n---\n\n\u003ctable\u003e\n\u003ctr\u003e\n\u003ctd width=\"50%\" valign=\"top\"\u003e\n\n#### 📱 **MobileNetV4**\n![ECCV](https://img.shields.io/badge/ECCV-2024-red?style=flat-square\u0026logo=adobeacrobatreader)\n![Mobile](https://img.shields.io/badge/Platform-Mobile-blue?style=flat-square\u0026logo=android)\n\n\u003e 🌐 Universal efficient architecture for mobile ecosystem\n\n**🎨 Innovations:**\n- 🔹 Universal Inverted Bottleneck (UIB) block\n- ⚡ Mobile MQA attention (**39% speedup**)\n- 🎯 Optimized NAS recipe\n- 🏆 **87% ImageNet accuracy** @ 3.8ms (Pixel 8 EdgeTPU)\n\n**📚 Resources:**\n- 📄 [MobileNetV4 Paper (Springer)](https://link.springer.com/chapter/10.1007/978-3-031-73661-2_5)\n- 🔬 [Google Research](https://syncedreview.com/2024/04/18/87-imagenet-accuracy-3-8ms-latency-googles-mobilenetv4-redefines-on-device-mobile-vision/)\n\n\u003c/td\u003e\n\u003ctd width=\"50%\" valign=\"top\"\u003e\n\n#### ⚡ **EfficientViT**\n![ViT](https://img.shields.io/badge/Type-Vision_Transformer-purple?style=flat-square\u0026logo=lightning)\n![2024](https://img.shields.io/badge/Year-2024-green?style=flat-square)\n\n\u003e 🧠 Lightweight multi-scale attention for high-resolution tasks\n\n**✨ Features:**\n- 🔸 Memory-efficient Vision Transformer\n- 🔸 Cascaded group attention\n- 🔸 Dense prediction tasks optimized\n- 🔸 High-resolution image processing\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/table\u003e\n\n---\n\n\u003cdiv align=\"center\"\u003e\n\n## 🤖 **Small Language Models (SLMs) for Edge**\n\n![LLM](https://img.shields.io/badge/Small_Language-Models-FF6B6B?style=for-the-badge\u0026logo=openai\u0026logoColor=white)\n![Edge](https://img.shields.io/badge/Edge-Deployment-4ECDC4?style=for-the-badge\u0026logo=raspberrypi\u0026logoColor=white)\n\n\u003c/div\u003e\n\n---\n\n\u003ctable\u003e\n\u003ctr\u003e\n\u003ctd width=\"50%\"\u003e\n\n### 🧠 **Microsoft Phi-3**\n![Microsoft](https://img.shields.io/badge/Microsoft-Phi--3-0078D4?style=for-the-badge\u0026logo=microsoft)\n\n**📊 Variants:**\n```yaml\nModel: Phi-3-mini\nParameters: 3.8B\nContext: Up to 128K tokens\nDeployment: GPU, CPU, Mobile\nStatus: ✅ Production Ready\n```\n\n**🎯 Optimized For:**\n- 💻 GPU acceleration\n- 🖥️ CPU inference\n- 📱 Mobile deployment\n\n**🔗 Resources:**\n- [Phi-3 Overview](https://datasciencedojo.com/blog/small-language-models-phi-3/)\n\n\u003c/td\u003e\n\u003ctd width=\"50%\"\u003e\n\n### 🦙 **TinyLlama**\n![TinyLlama](https://img.shields.io/badge/TinyLlama-1.1B-orange?style=for-the-badge\u0026logo=meta)\n\n**📊 Specifications:**\n```yaml\nParameters: 1.1B\nTarget: Mobile/Edge devices\nPerformance: High for size class\nYear: 2024\nStatus: ✅ Active\n```\n\n**✨ Highlights:**\n- 🔸 Compact architecture\n- 🔸 Edge-optimized\n- 🔸 Strong performance/size ratio\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"50%\"\u003e\n\n### 🌟 **Google Gemini Nano**\n![Google](https://img.shields.io/badge/Google-Gemini_Nano-4285F4?style=for-the-badge\u0026logo=google)\n\n**📱 On-device AI for Smartphones**\n\n**Variants:**\n- 📊 **1.8B** parameters (lightweight)\n- 📊 **3.25B** parameters (standard)\n\n**🎯 Capabilities:**\n- ✅ Context-aware reasoning\n- ✅ Real-time translation\n- ✅ Text summarization\n- ✅ Edge-optimized for phones/IoT\n\n\u003c/td\u003e\n\u003ctd width=\"50%\"\u003e\n\n### 🦙 **Meta Llama 3.2**\n![Meta](https://img.shields.io/badge/Meta-Llama_3.2-0668E1?style=for-the-badge\u0026logo=meta)\n\n**🖼️ Edge AI \u0026 Vision Capabilities**\n\n**Features:**\n- ⚡ Edge deployment optimized\n- 👁️ Vision-language capabilities\n- 📱 Mobile-friendly variants\n- 🔥 Latest architecture\n\n**🔗 Resources:**\n- [Llama 3.2 Announcement](https://ai.meta.com/blog/llama-3-2-connect-2024-vision-edge-mobile-devices/)\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/table\u003e\n\n---\n\n### 📷 **MobileVLM**\n![VLM](https://img.shields.io/badge/Vision--Language-Model-success?style=for-the-badge\u0026logo=youtube)\n\n\u003e 🎨 Efficient vision-language model for mobile devices\n\n**Specifications:**\n- 🔹 **mobileLLaMA**: 2.7B parameters\n- 🔹 Trained from scratch on open datasets\n- 🔹 Fully optimized for mobile deployment\n- 🔹 Vision + Language capabilities\n\n---\n\n\u003cdiv align=\"center\"\u003e\n\n## ⚡ **State Space Models - Efficient Transformers**\n\n![SSM](https://img.shields.io/badge/State_Space-Models-blueviolet?style=for-the-badge\u0026logo=lightning\u0026logoColor=white)\n![Efficiency](https://img.shields.io/badge/5x-Faster_Than_Transformers-gold?style=for-the-badge\u0026logo=speedtest)\n\n\u003c/div\u003e\n\n---\n\n\u003ctable\u003e\n\u003ctr\u003e\n\u003ctd width=\"50%\"\u003e\n\n### 🐍 **Mamba**\n![Mamba](https://img.shields.io/badge/Mamba-SSM-green?style=for-the-badge\u0026logo=python)\n\n\u003e ⚡ Linear-time sequence modeling with selective state spaces\n\n**🚀 Performance Highlights:**\n\n| Metric | Performance |\n|--------|-------------|\n| Throughput | **5x higher** than Transformers |\n| Scaling | **Linear** in sequence length |\n| Comparison | Mamba-3B \u003e Transformers (same size) |\n| Power | Matches Transformers 2x its size |\n\n**📊 Advantages:**\n```diff\n+ ✅ Linear time complexity\n+ ✅ 5x throughput improvement\n+ ✅ Efficient long sequences\n+ ✅ Lower memory footprint\n- ❌ Newer architecture (less tested)\n```\n\n**📚 Resources:**\n- 📄 [Mamba Paper](https://arxiv.org/abs/2312.00752)\n- 💻 [Mamba GitHub](https://github.com/state-spaces/mamba)\n- 📖 [Mamba Survey](https://arxiv.org/html/2408.01129v1)\n\n\u003c/td\u003e\n\u003ctd width=\"50%\"\u003e\n\n### 📱 **eMamba**\n![eMamba](https://img.shields.io/badge/eMamba-Edge_Optimized-orange?style=for-the-badge\u0026logo=raspberry-pi)\n\n\u003e 🔧 Edge-optimized Mamba acceleration framework\n\n**✨ Features:**\n```yaml\nDesign: End-to-end hardware acceleration\nTarget: Edge platforms\nComplexity: Linear time\nStatus: 2024 Release\n```\n\n**🎯 Optimizations:**\n- 🔹 Hardware-aware design\n- 🔹 Edge platform specific\n- 🔹 Leverages linear complexity\n- 🔹 Memory efficient\n\n**📚 Resources:**\n- 📄 [eMamba Paper](https://arxiv.org/html/2508.10370)\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/table\u003e\n\n---\n\n\u003cdiv align=\"center\"\u003e\n\n## 🚀 **Inference Frameworks \u0026 Runtimes**\n\n![Inference](https://img.shields.io/badge/High_Performance-Inference-FF6B6B?style=for-the-badge\u0026logo=nvidia\u0026logoColor=white)\n![Runtime](https://img.shields.io/badge/Runtime-Optimization-00D9FF?style=for-the-badge\u0026logo=apache\u0026logoColor=white)\n\n\u003c/div\u003e\n\n---\n\n\u003ctable\u003e\n\u003ctr\u003e\n\u003ctd width=\"50%\"\u003e\n\n### ⚡ **TensorRT-LLM**\n![NVIDIA](https://img.shields.io/badge/NVIDIA-TensorRT--LLM-76B900?style=for-the-badge\u0026logo=nvidia)\n\n\u003e 🏆 High-performance LLM inference on NVIDIA GPUs\n\n**📊 Performance:**\n```diff\n+ 70% faster than llama.cpp on RTX 4090\n+ State-of-the-art optimizations\n+ Quality maintained across precisions\n```\n\n**✨ Features:**\n- 🔸 Python \u0026 C++ API\n- 🔸 Multi-precision support\n- 🔸 Advanced kernel optimization\n- 🔸 Production-grade quality\n\n**🔗 Resources:**\n- [TensorRT-LLM GitHub](https://github.com/NVIDIA/TensorRT-LLM)\n- [Deployment Guide](https://towardsdatascience.com/deploying-llms-into-production-using-tensorrt-llm-ed36e620dac4/)\n\n\u003c/td\u003e\n\u003ctd width=\"50%\"\u003e\n\n### 📄 **vLLM**\n![vLLM](https://img.shields.io/badge/UC_Berkeley-vLLM-003262?style=for-the-badge\u0026logo=databricks)\n\n\u003e 💡 High-throughput LLM serving with PagedAttention\n\n**🎯 Innovations:**\n- ⚡ PagedAttention memory management\n- 🔸 Optimized KV cache handling\n- 🌐 Multi-platform support\n\n**🖥️ Supported Hardware:**\n```yaml\nAMD: GPU support\nGoogle: TPU support\nAWS: Inferentia support\nBase: PyTorch\n```\n\n**🔗 Resources:**\n- [vLLM vs TensorRT-LLM](https://northflank.com/blog/vllm-vs-tensorrt-llm-and-how-to-run-them)\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"50%\"\u003e\n\n### 🦙 **ExecuTorch**\n![Meta](https://img.shields.io/badge/Meta-ExecuTorch-0668E1?style=for-the-badge\u0026logo=meta)\n\n\u003e 📱 Efficient LLM execution on edge devices\n\n**Features:**\n- 🔹 Lightweight edge runtime\n- 🔹 Static memory planning\n- 🔹 Multi-platform support\n- 🔹 TorchAO quantization\n\n**💻 Hardware Support:**\n- ✅ CPU\n- ✅ GPU\n- ✅ AI Accelerators\n- ✅ Mobile devices\n\n**🔗 Resources:**\n- [PyTorch Conference 2024](https://www.infoq.com/news/2024/09/pytorch-conference-2024/)\n\n\u003c/td\u003e\n\u003ctd width=\"50%\"\u003e\n\n### 💻 **llama.cpp**\n![llama.cpp](https://img.shields.io/badge/llama.cpp-CPU_Optimized-green?style=for-the-badge\u0026logo=cplusplus)\n\n\u003e ⚡ CPU-optimized LLM inference\n\n**Advantages:**\n```diff\n+ ✅ Lower memory usage\n+ ✅ No GPU required\n+ ✅ Fast generation\n+ ✅ Cross-platform\n+ ✅ Wide model support\n```\n\n**🔗 Comparison:**\n- [vLLM vs Ollama vs llama.cpp vs TGI vs TensorRT-LLM](https://itecsonline.com/post/vllm-vs-ollama-vs-llama.cpp-vs-tgi-vs-tensort)\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/table\u003e\n\n---\n\n\u003cdiv align=\"center\"\u003e\n\n## 🔧 **Model Compression \u0026 Optimization**\n\n![Compression](https://img.shields.io/badge/Model-Compression-FF6B6B?style=for-the-badge\u0026logo=semanticscholar\u0026logoColor=white)\n![Quantization](https://img.shields.io/badge/Quantization-4bit__8bit-4ECDC4?style=for-the-badge\u0026logo=hackthebox\u0026logoColor=white)\n\n\u003c/div\u003e\n\n---\n\n### 📉 **Advanced Quantization Techniques**\n\n\u003ctable\u003e\n\u003ctr\u003e\n\u003ctd width=\"33%\"\u003e\n\n#### 🏆 **AWQ**\n![Award](https://img.shields.io/badge/MLSys_2024-Best_Paper-gold?style=flat-square\u0026logo=adobeacrobatreader)\n\n**Activation-aware Weight Quantization**\n\n\u003e 🎯 MIT HAN Lab Innovation\n\n**Key Concept:**\n```python\n# Not all weights are equal!\nif is_salient(weight):\n    skip_quantization()\nelse:\n    quantize_weight()\n```\n\n**Features:**\n- ⚡ Protects critical weights\n- 🎯 Activation-aware\n- 🔥 State-of-the-art results\n\n**🔗 Resources:**\n- [AWQ GitHub](https://github.com/mit-han-lab/llm-awq)\n- [MIT HAN Lab](https://hanlab.mit.edu/)\n\n\u003c/td\u003e\n\u003ctd width=\"33%\"\u003e\n\n#### 💎 **GPTQ**\n![GPTQ](https://img.shields.io/badge/GPTQ-4bit-blue?style=flat-square\u0026logo=lightning)\n\n**GPU-Focused Quantization**\n\n**Features:**\n- 🔸 Row-wise quantization\n- 🔸 Hessian optimization\n- 🔸 GPU inference focused\n- 🔸 175B models supported\n\n**Achievements:**\n```yaml\nModels: BLOOM, OPT-175B\nPrecision: 4-bit\nPlatform: GPU optimized\n```\n\n\u003c/td\u003e\n\u003ctd width=\"33%\"\u003e\n\n#### 🔬 **QLoRA**\n![QLoRA](https://img.shields.io/badge/QLoRA-Fine--Tuning-purple?style=flat-square\u0026logo=pytorch)\n\n**Efficient Fine-tuning**\n\n**Innovations:**\n- ✨ 4-bit NormalFloat (NF4)\n- ✨ Double quantization\n- ✨ LoRA adapters\n- ✨ Single GPU fine-tuning\n\n**Capability:**\n```diff\n+ Fine-tune 65B model\n+ On single GPU\n+ Maintain quality\n```\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/table\u003e\n\n---\n\n#### 🆕 **Unsloth Dynamic 4-bit**\n![Latest](https://img.shields.io/badge/Release-December_2024-brightgreen?style=for-the-badge\u0026logo=github)\n\n\u003e 🔥 Latest quantization innovation\n\n**Features:**\n- Built on BitsandBytes\n- Dynamic parameter quantization\n- Per-parameter optimization\n\n**📚 Comprehensive Guides:**\n- 📖 [Quantization Comparison](https://generativeai.pub/practical-guide-of-llm-quantization-gptq-awq-bitsandbytes-and-unsloth-bdeaa2c0bbf6)\n- 📊 [GPTQ vs GGUF vs AWQ](https://newsletter.maartengrootendorst.com/p/which-quantization-method-is-right)\n\n---\n\n### 🔬 **Neural Architecture Search (NAS)**\n\n\u003cdiv align=\"center\"\u003e\n\n![NAS](https://img.shields.io/badge/Neural_Architecture-Search-blueviolet?style=for-the-badge\u0026logo=pytorch)\n\n\u003c/div\u003e\n\n\u003e 🤖 Automate neural network architecture design\n\n#### 🎯 **Once-for-All (OFA)**\n\n**Concept:** Train once, deploy everywhere\n\n```mermaid\ngraph TD\n    A[🌐 Supernet Training] --\u003e B[📦 Weight Sharing]\n    B --\u003e C[📱 Mobile]\n    B --\u003e D[💻 Desktop]\n    B --\u003e E[⚡ Edge]\n    style A fill:#e1f5ff\n    style B fill:#ffe1f5\n    style C fill:#f5ffe1\n    style D fill:#ffe1e1\n    style E fill:#ffd700\n```\n\n**Features:**\n- 🔹 Weight-sharing supernetwork\n- 🔹 Represents any architecture in search space\n- 🔹 Massive computational savings\n- 🔹 Applied to ImageNet with ProxylessNAS \u0026 MobileNetV3\n\n**🔗 Resources:**\n- [NAS Overview](https://www.automl.org/nas-overview/)\n- [MIT HAN Lab NAS](https://hanlab.mit.edu/techniques/nas)\n\n---\n\n### 🎓 **Knowledge Distillation \u0026 Pruning**\n\n\u003ctable\u003e\n\u003ctr\u003e\n\u003ctd width=\"50%\"\u003e\n\n#### 🔬 **TinyBERT**\n![TinyBERT](https://img.shields.io/badge/TinyBERT-7.5x_Smaller-success?style=for-the-badge\u0026logo=semanticscholar)\n\n\u003e 📚 Two-stage distillation approach\n\n**Performance Metrics:**\n```yaml\nAccuracy: 96.8% of BERT-base\nSize: 7.5x smaller (4 layers)\nEnergy: Lowest variability (0.1032 kWh SD)\nStages: Task-agnostic + Task-specific\n```\n\n**Advantages:**\n- ✅ Dual-stage distillation\n- ✅ Ultra-low energy variability\n- ✅ Compact architecture\n- ✅ High performance retention\n\n\u003c/td\u003e\n\u003ctd width=\"50%\"\u003e\n\n#### 📖 **DistilBERT**\n![DistilBERT](https://img.shields.io/badge/DistilBERT-40%25_Smaller-blue?style=for-the-badge\u0026logo=huggingface)\n\n\u003e ⚡ Single-phase task-agnostic distillation\n\n**Performance Metrics:**\n```yaml\nAccuracy: 97% of BERT\nSize Reduction: 40% smaller\nSpeed: 60% faster\nUse Case: General-purpose\n```\n\n**Recent Research (2025):**\n- 🔸 32% energy reduction with pruning\n- 🔸 Iterative distillation + adaptive pruning\n- 🔸 Nature Scientific Reports\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/table\u003e\n\n**📚 Resources:**\n- [Nature Scientific Reports 2025](https://www.nature.com/articles/s41598-025-07821-w)\n- [DistilBERT Medium](https://medium.com/huggingface/distilbert-8cf3380435b5)\n\n---\n\n\u003cdiv align=\"center\"\u003e\n\n## 🎯 **TinyML \u0026 MCU-specific Advances**\n\n![TinyML](https://img.shields.io/badge/TinyML-Microcontrollers-FF6B6B?style=for-the-badge\u0026logo=arduino\u0026logoColor=white)\n![MIT](https://img.shields.io/badge/MIT-HAN_Lab-A31F34?style=for-the-badge\u0026logo=mit\u0026logoColor=white)\n\n\u003c/div\u003e\n\n---\n\n### 🧠 **MCUNet Series** - MIT HAN Lab\n\n\u003ctable\u003e\n\u003ctr\u003e\n\u003ctd width=\"33%\"\u003e\n\n#### 📱 **MCUNetV1**\n![V1](https://img.shields.io/badge/Version-1.0-blue?style=flat-square)\n\n**Foundation:**\n- 🔸 Neural architecture for MCUs\n- 🔸 Co-designed model + inference engine\n- 🔸 Ultra-low memory footprint\n\n\u003c/td\u003e\n\u003ctd width=\"33%\"\u003e\n\n#### 🚀 **MCUNetV2**\n![V2](https://img.shields.io/badge/Version-2.0-green?style=flat-square)\n\n**Achievements:**\n```yaml\nImageNet: 71.8% accuracy\nVisual Wake: \u003e90% (32kB SRAM)\nCapability: Object detection\nPlatform: Tiny devices\n```\n\n\u003c/td\u003e\n\u003ctd width=\"33%\"\u003e\n\n#### ⚡ **MCUNetV3**\n![V3](https://img.shields.io/badge/Version-3.0-orange?style=flat-square)\n\n**Latest:**\n- 🔸 Enhanced efficiency\n- 🔸 State-of-the-art MCU AI\n- 🔸 Production ready\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/table\u003e\n\n---\n\n#### 🎓 **Additional MCU Tools**\n\n\u003ctable\u003e\n\u003ctr\u003e\n\u003ctd width=\"50%\"\u003e\n\n**🔧 TinyTL**\n- Tiny transfer learning for MCUs\n- On-device learning capabilities\n- Minimal resource overhead\n\n\u003c/td\u003e\n\u003ctd width=\"50%\"\u003e\n\n**⚙️ PockEngine**\n- Inference engine optimization\n- MCU-specific acceleration\n- Memory-efficient execution\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/table\u003e\n\n**📚 Resources:**\n- 🌐 [MCUNet Official](https://mcunet.mit.edu/)\n- 💻 [MCUNet GitHub](https://github.com/mit-han-lab/mcunet)\n- 📖 [TinyML Projects](https://hanlab.mit.edu/projects/tinyml)\n\n---\n\n### 🔬 **TinyDL (Tiny Deep Learning)**\n\n![TinyDL](https://img.shields.io/badge/TinyDL-2024-blueviolet?style=for-the-badge\u0026logo=tensorflow)\n\n\u003e 🎯 Evolution from TinyML to deep learning on edge\n\n**Focus Areas:**\n- 🔹 Deep learning on ultra-constrained hardware\n- 🔹 Power consumption in **mW range**\n- 🔹 On-device sensor analytics\n- 🔹 Real-time inference\n\n**📄 Resources:**\n- [TinyDL Survey](https://arxiv.org/html/2506.18927v1)\n\n---\n\n\u003cdiv align=\"center\"\u003e\n\n## 🔩 **Hardware Acceleration \u0026 Platforms**\n\n![Hardware](https://img.shields.io/badge/Hardware-Acceleration-gold?style=for-the-badge\u0026logo=nvidia\u0026logoColor=black)\n![Edge](https://img.shields.io/badge/Edge-Devices-4ECDC4?style=for-the-badge\u0026logo=raspberrypi\u0026logoColor=white)\n\n\u003c/div\u003e\n\n---\n\n### 🖥️ **Edge AI Platforms**\n\n\u003ctable\u003e\n\u003ctr\u003e\n\u003ctd width=\"50%\"\u003e\n\n#### 🟢 **NVIDIA Jetson Orin Nano Super**\n![NVIDIA](https://img.shields.io/badge/NVIDIA-Jetson-76B900?style=for-the-badge\u0026logo=nvidia)\n\n**Specifications:**\n```yaml\nCompute: 67 INT8 TOPS\nPerformance: 1.7x vs previous Orin\nPrice: $249\nRelease: Late 2024\nStatus: ✅ Available\n```\n\n**Features:**\n- ⚡ Generative AI optimized\n- 🎯 Edge AI development kit\n- 💰 Affordable price point\n\n\u003c/td\u003e\n\u003ctd width=\"50%\"\u003e\n\n#### 🔷 **Edge TPU \u0026 Neural Accelerators**\n\n**Hardware Platforms:**\n\n![Google](https://img.shields.io/badge/Google-Edge_TPU-4285F4?style=flat-square\u0026logo=google)\n- Google Pixel EdgeTPU\n- Coral Dev Board\n\n![Apple](https://img.shields.io/badge/Apple-Neural_Engine-000000?style=flat-square\u0026logo=apple)\n- Apple Neural Engine\n- A-series chips\n\n![Generic](https://img.shields.io/badge/Generic-AI_Accelerators-orange?style=flat-square\u0026logo=sparkfun)\n- Specialized NPUs\n- Custom ASICs\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/table\u003e\n\n---\n\n### 📱 **Mobile Deployment Targets**\n\n\u003cdiv align=\"center\"\u003e\n\n| Platform | Architecture | Use Case |\n|----------|-------------|----------|\n| 🔧 **ARM CPUs** | ARM Cortex | General compute |\n| 📡 **Mobile DSPs** | Qualcomm/MediaTek | Signal processing |\n| 🎮 **Mobile GPUs** | Mali/Adreno | Graphics + AI |\n| 🧠 **NPUs** | Custom ASICs | Neural processing |\n\n\u003c/div\u003e\n\n---\n\n\u003cdiv align=\"center\"\u003e\n\n## 🛠️ **Implementation Resources \u0026 Tools**\n\n![ONNX](https://img.shields.io/badge/ONNX-Runtime-blue?style=for-the-badge\u0026logo=onnx\u0026logoColor=white)\n![TensorRT](https://img.shields.io/badge/TensorRT-NVIDIA-76B900?style=for-the-badge\u0026logo=nvidia\u0026logoColor=white)\n\n\u003c/div\u003e\n\n---\n\n### 🔷 **ONNX Runtime**\n\n\u003e Cross-platform inference with ONNX models\n\n\u003ctable\u003e\n\u003ctr\u003e\n\u003ctd width=\"50%\" valign=\"top\"\u003e\n\n#### 📚 **Documentation \u0026 Tutorials**\n- 📖 [ONNX Runtime C++ Inference](https://leimao.github.io/blog/ONNX-Runtime-CPP-Inference/)\n- 🐍 [PyTorch to ONNX Tutorial](https://pytorch.org/tutorials/advanced/super_resolution_with_onnxruntime.html)\n- 📝 [ONNX Registry Tutorial](https://pytorch.org/tutorials/beginner/onnx/onnx_registry_tutorial.html)\n- 🎓 [On-Device Training](https://onnxruntime.ai/docs/api/python/on_device_training/training_artifacts.html)\n\n#### 🔧 **Compatibility**\n- ⚙️ [ONNX Runtime Compatibility](https://onnxruntime.ai/docs/reference/compatibility.html)\n- 📋 [ONNX Versioning](https://github.com/onnx/onnx/blob/main/docs/Versioning.md)\n- 🚀 [CUDA Execution Provider](https://onnxruntime.ai/docs/execution-providers/CUDA-ExecutionProvider.html)\n\n\u003c/td\u003e\n\u003ctd width=\"50%\" valign=\"top\"\u003e\n\n#### 💻 **Example Implementations**\n- 🖼️ [C++ ResNet Console App](https://github.com/cassiebreviu/cpp-onnxruntime-resnet-console-app)\n- ⚡ [ONNX Runtime C++ Example](https://github.com/k2-gc/onnxruntime-cpp-example)\n- 🤖 [ONNX Runtime Android](https://github.com/Rohithkvsp/OnnxRuntimeAndorid)\n- 🎯 [ByteTrack ONNX Inference](https://github.com/ifzhang/ByteTrack/blob/main/deploy/ONNXRuntime/onnx_inference.py)\n\n#### 📦 **Model Repositories**\n- 🤗 [HuggingFace ONNX Models](https://huggingface.co/models?sort=trending\u0026search=onnx)\n- 🔧 [txtai ONNX Pipeline](https://neuml.github.io/txtai/pipeline/train/hfonnx/)\n- 📤 [Ultralytics Export](https://docs.ultralytics.com/modes/export/#arguments)\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/table\u003e\n\n---\n\n### 📉 **ONNX Runtime Quantization**\n\n![Quantization](https://img.shields.io/badge/Quantization-INT8_FP16-success?style=for-the-badge\u0026logo=semanticscholar)\n\n**Tools \u0026 Resources:**\n- 🔧 [Quantization Tools](https://github.com/microsoft/onnxruntime/tree/main/onnxruntime/python/tools/quantization)\n- 📊 [Float16 Optimization](https://onnxruntime.ai/docs/performance/model-optimizations/float16.html)\n- 💡 [Quantization Examples](https://github.com/microsoft/onnxruntime-inference-examples/tree/main/quantization)\n\n---\n\n### 🎯 **YOLO Implementations**\n\n\u003cdetails\u003e\n\u003csummary\u003e\u003cb\u003e🔥 Click to expand YOLO implementations\u003c/b\u003e\u003c/summary\u003e\n\n\u003cbr\u003e\n\n#### 🟣 **YOLO-NAS with ONNX**\n- 💻 [YOLO-NAS ONNXRuntime](https://github.com/jason-li-831202/YOLO-NAS-onnxruntime)\n\n#### 🟢 **YOLO + TensorRT** (Detection, Pose, Segmentation)\n- ⚡ [YOLOv8-TensorRT-CPP](https://github.com/cyrusbehr/YOLOv8-TensorRT-CPP)\n- 🔧 [TensorRT C++ API](https://github.com/cyrusbehr/tensorrt-cpp-api)\n- 🐍 [YOLOv8-TensorRT (Python + C++)](https://github.com/triple-Mu/YOLOv8-TensorRT)\n- 🤸 [YOLO Pose C++](https://github.com/mattiasbax/yolo-pose_cpp)\n- 📚 [TensorRT Samples](https://github.com/NVIDIA/TensorRT/tree/main/samples/trtexec)\n- 📺 [YOLOv8 TensorRT Tutorial](https://www.youtube.com/watch?v=Z0n5aLmcRHQ)\n\n#### 🔵 **YOLO + ONNXRuntime** (All Tasks)\n- 💻 [YOLOv8-ONNX-CPP](https://github.com/FourierMourier/yolov8-onnx-cpp/tree/main)\n- 🤸 [YOLOv8 Pose Implementation](https://github.com/mallumoSK/yolov8/blob/master/yolo/YoloPose.cpp)\n- ⚡ [YOLOv8 TensorRT Pose](https://github.com/triple-Mu/YOLOv8-TensorRT/blob/main/csrc/pose/normal/main.cpp)\n- 🔧 [YOLO-ONNXRuntime-CPP](https://github.com/Amyheart/yolo-onnxruntime-cpp)\n- 📷 [YOLOv8-OpenCV-ONNXRuntime-CPP](https://github.com/UNeedCryDear/yolov8-opencv-onnxruntime-cpp)\n- 📖 [Ultralytics YOLOv8 C++](https://github.com/ultralytics/ultralytics/tree/main/examples/YOLOv8-ONNXRuntime-CPP)\n- 🎯 [YOLOv6-OpenCV-ONNXRuntime](https://github.com/hpc203/yolov6-opencv-onnxruntime/tree/main)\n- 🏃 [YOLOv5 Pose OpenCV](https://github.com/hpc203/yolov5_pose_opencv)\n\n#### 🌐 **Community Resources**\n- 👨‍💻 [hpc203 Repositories](https://github.com/hpc203?tab=repositories)\n- 💬 [YOLO Issue Discussions](https://github.com/ultralytics/ultralytics/issues/1852)\n- 🐛 [YOLOv5 Fixed Bugs](https://github.com/ultralytics/yolov5/issues/916)\n- 🇨🇳 [Chinese Tutorial](https://zhuanlan.zhihu.com/p/466677699)\n- 📦 [ONNX Runtime Install Guide](https://velog.io/@dnchoi/ONNX-runtime-install)\n\n\u003c/details\u003e\n\n---\n\n### ⚡ **TensorRT**\n\n![TensorRT](https://img.shields.io/badge/TensorRT-Inference_Optimizer-76B900?style=for-the-badge\u0026logo=nvidia)\n\n\u003e 🚀 NVIDIA's high-performance deep learning inference optimizer\n\n**Resources:**\n- 🔧 [TensorRT Execution Provider](https://onnxruntime.ai/docs/execution-providers/TensorRT-ExecutionProvider.html#requirements)\n- 💾 [TensorRT Engine Cache](https://gitee.com/arnoldfychen/onnxruntime/blob/master/docs/execution_providers/TensorRT-ExecutionProvider.md#specify-tensorrt-engine-cache-path)\n\n---\n\n\u003cdiv align=\"center\"\u003e\n\n## 🌐 **Edge Deployment Frameworks**\n\n![Deployment](https://img.shields.io/badge/Edge-Deployment-FF6B6B?style=for-the-badge\u0026logo=kubernetes\u0026logoColor=white)\n![Frameworks](https://img.shields.io/badge/Frameworks-Multi--Platform-4ECDC4?style=for-the-badge\u0026logo=docker\u0026logoColor=white)\n\n\u003c/div\u003e\n\n---\n\n\u003ctable\u003e\n\u003ctr\u003e\n\u003ctd width=\"50%\"\u003e\n\n### 🚀 **FastDeploy** - PaddlePaddle\n![PaddlePaddle](https://img.shields.io/badge/PaddlePaddle-FastDeploy-blue?style=for-the-badge)\n\n\u003e 📦 Easy-to-use deployment toolbox for AI models\n\n**Resources:**\n- 💻 [FastDeploy GitHub](https://github.com/PaddlePaddle/FastDeploy)\n- 📥 [Prebuilt Libraries](https://github.com/PaddlePaddle/FastDeploy/blob/develop/docs/en/build_and_install/download_prebuilt_libraries.md)\n\n---\n\n### 💎 **DeepSparse \u0026 SparseML** - Neural Magic\n![Neural Magic](https://img.shields.io/badge/Neural_Magic-DeepSparse-purple?style=for-the-badge)\n\n\u003e 🖥️ CPU-optimized inference with sparsity\n\n**Features:**\n- ⚡ CPU inference acceleration\n- 🔸 Sparsity-aware optimization\n- 📊 YOLOv5 CPU benchmarks\n\n**Resources:**\n- 📈 [YOLOv5 CPU Benchmark](https://neuralmagic.com/blog/benchmark-yolov5-on-cpus-with-deepsparse/)\n- 💻 [SparseML GitHub](https://github.com/neuralmagic/sparseml/tree/main)\n- 🚀 [DeepSparse GitHub](https://github.com/neuralmagic/deepsparse)\n\n\u003c/td\u003e\n\u003ctd width=\"50%\"\u003e\n\n### 📱 **NCNN** - Tencent\n![Tencent](https://img.shields.io/badge/Tencent-NCNN-00D9FF?style=for-the-badge)\n\n\u003e 🎯 High-performance neural network inference for mobile\n\n**Resources:**\n- 💻 [NCNN GitHub](https://github.com/Tencent/ncnn)\n- 📖 [NCNN C++ Usage](https://github.com/Tencent/ncnn/blob/master/docs/how-to-use-and-FAQ/use-ncnn-with-alexnet.md)\n- 📱 [YoloMobile](https://github.com/wkt/YoloMobile)\n- ⭐ [Awesome NCNN](https://github.com/umitkacar/awesome-ncnn-collection)\n- 🔄 [Model Converter](https://convertmodel.com/)\n\n---\n\n### 🔧 **MACE** - Xiaomi\n![Xiaomi](https://img.shields.io/badge/Xiaomi-MACE-FF6900?style=for-the-badge)\n\n\u003e 🤖 Mobile AI Compute Engine\n\n**Resources:**\n- 💻 [MACE GitHub](https://github.com/xiaomi/mace)\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/table\u003e\n\n---\n\n### 🍎 **CoreML** - Apple\n\n![Apple](https://img.shields.io/badge/Apple-CoreML-000000?style=for-the-badge\u0026logo=apple)\n\n\u003e 🎨 Machine learning framework for iOS/macOS\n\n\u003cdetails\u003e\n\u003csummary\u003e\u003cb\u003e📦 Click to expand CoreML resources\u003c/b\u003e\u003c/summary\u003e\n\n\u003cbr\u003e\n\n#### 🎨 **Model Collections**\n- 🎯 [Semantic Segmentation CoreML](https://github.com/tucan9389/SemanticSegmentation-CoreML)\n- 📚 [CoreML Models Collection](https://github.com/john-rocky/CoreML-Models#u2net)\n- ⭐ [Awesome CoreML Models](https://github.com/likedan/Awesome-CoreML-Models)\n- 🧠 [Awesome CoreML Models 2](https://github.com/SwiftBrain/awesome-CoreML-models)\n- 🎬 [RobustVideoMatting](https://github.com/PeterL1n/RobustVideoMatting)\n\n#### 🛠️ **Tools \u0026 Documentation**\n- 🔄 [PyTorch to CoreML](https://coremltools.readme.io/docs/pytorch-conversion)\n- 🔧 [CoreML Helpers](https://github.com/hollance/CoreMLHelpers)\n- 📖 [Apple ML API](https://developer.apple.com/machine-learning/api/)\n- 📊 [CoreML Performance Tool](https://github.com/vladimir-chernykh/coreml-performance)\n\n#### 🎨 **Stable Diffusion on CoreML**\n- 🔬 [Apple ML-4M](https://github.com/apple/ml-4m/)\n- 🎯 [Apple ML Stable Diffusion](https://github.com/apple/ml-stable-diffusion)\n- 📦 [Stable Diffusion 2 Base](https://huggingface.co/stabilityai/stable-diffusion-2-base)\n- 🚀 [Stability AI SD](https://github.com/Stability-AI/stablediffusion)\n- 📚 [Stable Diffusion v1.4](https://huggingface.co/CompVis/stable-diffusion-v1-4)\n- 🎬 [RunwayML SD](https://github.com/runwayml/stable-diffusion)\n- 🖼️ [Automatic1111 WebUI](https://github.com/AUTOMATIC1111/stable-diffusion-webui)\n\n\u003c/details\u003e\n\n---\n\n\u003cdiv align=\"center\"\u003e\n\n## ⚙️ **Compilers \u0026 Low-Level Frameworks**\n\n![Compilers](https://img.shields.io/badge/Compilers-Low--Level-blueviolet?style=for-the-badge\u0026logo=llvm\u0026logoColor=white)\n![Optimization](https://img.shields.io/badge/Hardware-Optimization-gold?style=for-the-badge\u0026logo=arm\u0026logoColor=black)\n\n\u003c/div\u003e\n\n---\n\n\u003ctable\u003e\n\u003ctr\u003e\n\u003ctd width=\"50%\"\u003e\n\n### 🔧 **TVM** - Apache\n![TVM](https://img.shields.io/badge/Apache-TVM-D22128?style=for-the-badge\u0026logo=apache)\n\n\u003e 🎯 End-to-end deep learning compiler stack\n\n**Resources:**\n- [TVM GitHub](https://github.com/apache/tvm)\n\n---\n\n### 🔨 **LLVM**\n![LLVM](https://img.shields.io/badge/LLVM-Compiler-262D3A?style=for-the-badge\u0026logo=llvm)\n\n\u003e ⚙️ Compiler infrastructure project\n\n**Resources:**\n- [LLVM Project](https://github.com/llvm/llvm-project)\n\n---\n\n### ⚡ **XNNPack** - Google\n![Google](https://img.shields.io/badge/Google-XNNPack-4285F4?style=for-the-badge\u0026logo=google)\n\n\u003e 🚀 High-efficiency floating-point neural network operators\n\n**Resources:**\n- [XNNPack GitHub](https://github.com/google/XNNPACK)\n\n\u003c/td\u003e\n\u003ctd width=\"50%\"\u003e\n\n### 🔷 **ARM-NN**\n![ARM](https://img.shields.io/badge/ARM-NN-0091BD?style=for-the-badge\u0026logo=arm)\n\n\u003e 💪 Inference engine for ARM platforms\n\n**Resources:**\n- [ARM-NN GitHub](https://github.com/ARM-software/armnn)\n- [ARM-NN Tutorial](https://www.youtube.com/watch?v=QuNOaFLobSg)\n\n---\n\n### 🧠 **CMSIS-NN**\n![CMSIS](https://img.shields.io/badge/ARM-CMSIS--NN-00979D?style=for-the-badge\u0026logo=arm)\n\n\u003e 📱 Efficient neural network kernels for ARM Cortex-M\n\n**Resources:**\n- [CMSIS-NN GitHub](https://github.com/ARM-software/CMSIS_5)\n\n---\n\n### 📱 **Samsung ONE**\n![Samsung](https://img.shields.io/badge/Samsung-ONE-1428A0?style=for-the-badge\u0026logo=samsung)\n\n\u003e 🔧 On-device Neural Engine compiler\n\n**Resources:**\n- [ONE GitHub](https://github.com/Samsung/ONE)\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/table\u003e\n\n---\n\n\u003cdiv align=\"center\"\u003e\n\n## 💼 **Industry \u0026 Commercial Solutions**\n\n![Industry](https://img.shields.io/badge/Industry-Solutions-FF6B6B?style=for-the-badge\u0026logo=enterprisedb\u0026logoColor=white)\n\n\u003c/div\u003e\n\n---\n\n### 🚀 **Deeplite**\n\n![Deeplite](https://img.shields.io/badge/Deeplite-AI_Optimizer-4ECDC4?style=for-the-badge)\n\n\u003e 🎯 AI-Driven Optimizer for Deep Neural Networks\n\n**Focus:**\n\n\u003ctable\u003e\n\u003ctr\u003e\n\u003ctd width=\"20%\" align=\"center\"\u003e⚡\u003cbr\u003e\u003cb\u003eFaster\u003cbr\u003eInference\u003c/b\u003e\u003c/td\u003e\n\u003ctd width=\"20%\" align=\"center\"\u003e📦\u003cbr\u003e\u003cb\u003eSmaller\u003cbr\u003eModels\u003c/b\u003e\u003c/td\u003e\n\u003ctd width=\"20%\" align=\"center\"\u003e🔋\u003cbr\u003e\u003cb\u003eEnergy\u003cbr\u003eEfficient\u003c/b\u003e\u003c/td\u003e\n\u003ctd width=\"20%\" align=\"center\"\u003e☁️\u003cbr\u003e\u003cb\u003eCloud to\u003cbr\u003eEdge\u003c/b\u003e\u003c/td\u003e\n\u003ctd width=\"20%\" align=\"center\"\u003e🎯\u003cbr\u003e\u003cb\u003eMaintain\u003cbr\u003eAccuracy\u003c/b\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/table\u003e\n\n**🔗 Resources:**\n- [Deeplite Website](https://www.deeplite.ai/)\n\n---\n\n\u003cdiv align=\"center\"\u003e\n\n## 🔧 **Utility Frameworks \u0026 Tools**\n\n![Tools](https://img.shields.io/badge/Utility-Tools-00D9FF?style=for-the-badge\u0026logo=hackthebox\u0026logoColor=white)\n\n\u003c/div\u003e\n\n---\n\n\u003ctable\u003e\n\u003ctr\u003e\n\u003ctd width=\"50%\"\u003e\n\n### 👁️ **OpenCV**\n![OpenCV](https://img.shields.io/badge/OpenCV-Computer_Vision-5C3EE8?style=for-the-badge\u0026logo=opencv)\n\n\u003e 📷 Computer vision library with C++ support\n\n**Resources:**\n- 📺 [OpenCV C++ Playlist](https://www.youtube.com/playlist?list=PLUTbi0GOQwghR9db9p6yHqwvzc989q_mu)\n- 🔨 [Build OpenCV C++](https://gist.github.com/raulqf/f42c718a658cddc16f9df07ecc627be7)\n\n\u003c/td\u003e\n\u003ctd width=\"50%\"\u003e\n\n### 🎬 **VQRF** - Video Compression\n![VQRF](https://img.shields.io/badge/VQRF-Video_Compression-red?style=for-the-badge\u0026logo=youtube)\n\n\u003e 📹 Vector Quantized Radiance Fields\n\n**Resources:**\n- [VQRF GitHub](https://github.com/AlgoHunt/VQRF)\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/table\u003e\n\n---\n\n\u003cdiv align=\"center\"\u003e\n\n## 🖼️ **Additional Model Architectures**\n\n![Models](https://img.shields.io/badge/Model-Architectures-blueviolet?style=for-the-badge\u0026logo=pytorch\u0026logoColor=white)\n\n\u003c/div\u003e\n\n---\n\n\u003ctable\u003e\n\u003ctr\u003e\n\u003ctd width=\"50%\"\u003e\n\n### 🎯 **PP-PicoDet**\n![PicoDet](https://img.shields.io/badge/PaddlePaddle-PicoDet-blue?style=for-the-badge)\n\n\u003e 📱 Lightweight real-time object detector for mobile\n\n**Resources:**\n- [PP-PicoDet Paper](https://arxiv.org/pdf/2111.00902.pdf)\n\n\u003c/td\u003e\n\u003ctd width=\"50%\"\u003e\n\n### 🔬 **EtinyNet**\n![EtinyNet](https://img.shields.io/badge/EtinyNet-TinyML-orange?style=for-the-badge\u0026logo=arduino)\n\n\u003e 🎯 Extremely tiny network for TinyML\n\n**Resources:**\n- [EtinyNet GitHub](https://github.com/aztc/EtinyNet)\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/table\u003e\n\n\u003cdiv align=\"center\"\u003e\n\n![TinyML Architecture](./tinyML.png)\n\n\u003c/div\u003e\n\n---\n\n\u003cdiv align=\"center\"\u003e\n\n## 🧠 **Computing Architectures \u0026 APIs**\n\n![Computing](https://img.shields.io/badge/Computing-Architectures-gold?style=for-the-badge\u0026logo=nvidia\u0026logoColor=black)\n\n\u003c/div\u003e\n\n---\n\n\u003ctable align=\"center\"\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\" width=\"16.66%\"\u003e\n\n![ARM](https://img.shields.io/badge/ARM-0091BD?style=for-the-badge\u0026logo=arm\u0026logoColor=white)\n\n**Mobile \u0026\u003cbr\u003eEmbedded**\n\n\u003c/td\u003e\n\u003ctd align=\"center\" width=\"16.66%\"\u003e\n\n![RISC-V](https://img.shields.io/badge/RISC--V-283272?style=for-the-badge\u0026logo=riscv\u0026logoColor=white)\n\n**Open-Source\u003cbr\u003eISA**\n\n\u003c/td\u003e\n\u003ctd align=\"center\" width=\"16.66%\"\u003e\n\n![CUDA](https://img.shields.io/badge/CUDA-76B900?style=for-the-badge\u0026logo=nvidia\u0026logoColor=white)\n\n**NVIDIA\u003cbr\u003eGPU**\n\n\u003c/td\u003e\n\u003ctd align=\"center\" width=\"16.66%\"\u003e\n\n![Metal](https://img.shields.io/badge/Metal-000000?style=for-the-badge\u0026logo=apple\u0026logoColor=white)\n\n**Apple\u003cbr\u003eGPU**\n\n\u003c/td\u003e\n\u003ctd align=\"center\" width=\"16.66%\"\u003e\n\n![OpenCL](https://img.shields.io/badge/OpenCL-721412?style=for-the-badge\u0026logo=opencl\u0026logoColor=white)\n\n**Cross-\u003cbr\u003ePlatform**\n\n\u003c/td\u003e\n\u003ctd align=\"center\" width=\"16.66%\"\u003e\n\n![Vulkan](https://img.shields.io/badge/Vulkan-AC162C?style=for-the-badge\u0026logo=vulkan\u0026logoColor=white)\n\n**Graphics \u0026\u003cbr\u003eCompute**\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/table\u003e\n\n---\n\n\u003cdiv align=\"center\"\u003e\n\n## 📚 **Research Papers \u0026 Academic Resources**\n\n![Research](https://img.shields.io/badge/Research-Papers-FF6B6B?style=for-the-badge\u0026logo=semanticscholar\u0026logoColor=white)\n![2024-2025](https://img.shields.io/badge/Years-2024--2025-4ECDC4?style=for-the-badge\u0026logo=academiasquare\u0026logoColor=white)\n\n\u003c/div\u003e\n\n---\n\n### 📖 **Foundational Surveys (2024-2025)**\n\n\u003cdetails open\u003e\n\u003csummary\u003e\u003cb\u003e🔍 Click to expand research papers\u003c/b\u003e\u003c/summary\u003e\n\n\u003cbr\u003e\n\n#### 🌐 **Edge Computing \u0026 Deep Learning**\n- 📄 [Deep Learning With Edge Computing: A Review](https://www.cs.ucr.edu/~jiasi/pub/deep_edge_review.pdf)\n- 📄 [Convergence of Edge Computing and Deep Learning](https://arxiv.org/pdf/1907.08349.pdf)\n- 📄 [Machine Learning at the Network Edge](https://arxiv.org/pdf/1908.00080.pdf)\n- 📄 [Edge Deep Learning in CV \u0026 Medical Diagnostics](https://link.springer.com/article/10.1007/s10462-024-11033-5)\n\n#### 🔬 **TinyML Specific**\n- 📄 [From Tiny ML to Tiny DL: A Survey (2024)](https://arxiv.org/html/2506.18927v1)\n- 📄 [EtinyNet: Extremely Tiny Network](https://ojs.aaai.org/index.php/AAAI/article/download/20387/version/18684/20146)\n- 📄 [Ultra-low Power TinyML System](https://arxiv.org/pdf/2207.04663.pdf)\n\n#### ⚡ **State Space Models \u0026 Efficient Architectures**\n- 📄 [Mamba: Linear-Time Sequence Modeling](https://arxiv.org/abs/2312.00752)\n- 📄 [Mamba-360: Survey of SSMs](https://arxiv.org/html/2404.16112v1)\n- 📄 [eMamba: Efficient Edge Acceleration](https://arxiv.org/html/2508.10370)\n\n#### 👁️ **Vision Models**\n- 📄 [MobileNetV4 (ECCV 2024)](https://link.springer.com/chapter/10.1007/978-3-031-73661-2_5)\n- 📄 [ViT for Mobile/Edge Devices](https://link.springer.com/article/10.1007/s00530-024-01312-0)\n- 📄 [YOLO Evolution: v5 to YOLO26](https://arxiv.org/html/2510.09653v2)\n- 📄 [YOLOv10: Real-Time Detection](https://arxiv.org/pdf/2405.14458)\n\n#### 🔧 **Model Compression \u0026 Optimization**\n- 📄 [Model Compression for Carbon Efficient AI (2025)](https://www.nature.com/articles/s41598-025-07821-w)\n- 📄 [NAS Systematic Review (2024)](https://link.springer.com/article/10.1007/s10462-024-11058-w)\n- 📄 [Advances in Neural Architecture Search](https://academic.oup.com/nsr/article/11/8/nwae282/7740455)\n\n#### 📚 **Collections**\n- ⭐ [Awesome Embedded and Mobile Deep Learning](https://github.com/csarron/awesome-emdl/blob/master/README.md)\n\n\u003c/details\u003e\n\n---\n\n\u003cdiv align=\"center\"\u003e\n\n## 🎓 **Contributing \u0026 Community**\n\n![Community](https://img.shields.io/badge/Community-Welcome-success?style=for-the-badge\u0026logo=github\u0026logoColor=white)\n![Contributions](https://img.shields.io/badge/Contributions-Open-blue?style=for-the-badge\u0026logo=githubactions\u0026logoColor=white)\n\n\u003c/div\u003e\n\n---\n\n\u003cdiv align=\"center\"\u003e\n\nThis repository serves as a **comprehensive resource** for AI edge computing and TinyML practitioners.\n\n**Contributions, updates, and corrections are welcome!** 🚀\n\n---\n\n### 📊 **Repository Stats**\n\n![Last Commit](https://img.shields.io/github/last-commit/umitkacar/ai-edge-computing-tiny-embedded?style=for-the-badge)\n![Contributors](https://img.shields.io/github/contributors/umitkacar/ai-edge-computing-tiny-embedded?style=for-the-badge)\n![Issues](https://img.shields.io/github/issues/umitkacar/ai-edge-computing-tiny-embedded?style=for-the-badge)\n\n---\n\n### 🏷️ **Keywords**\n\n`TinyML` • `Edge AI` • `Embedded ML` • `Model Compression` • `Quantization` • `Neural Architecture Search` • `YOLO` • `MobileNet` • `Transformer` • `State Space Models` • `ONNX Runtime` • `TensorRT` • `Inference Optimization` • `MCU` • `IoT` • `Real-Time AI`\n\n---\n\n### 📅 **Last Updated**\n**January 2025**\n\n---\n\n\u003cimg src=\"https://capsule-render.vercel.app/api?type=waving\u0026color=gradient\u0026customColorList=6,11,20\u0026height=150\u0026section=footer\u0026text=Thank%20You!\u0026fontSize=42\u0026fontColor=fff\u0026animation=twinkling\u0026fontAlignY=72\"/\u003e\n\n\u003c/div\u003e\n","projects_url":"https://awesome.ecosyste.ms/api/v1/lists/umitkacar%2Fawesome-tinyml/projects"}