[{"data":1,"prerenderedAt":27},["ShallowReactive",2],{"chapter:teenygrad\u002Fintroduction.json":3},{"project":4,"route":5,"title":6,"titleHtml":6,"navTitle":6,"part":7,"sourcePath":8,"editUrl":9,"html":10,"toc":11,"hasMermaid":22,"prev":7,"next":23},"teenygrad","\u002Fteenygrad","Introduction",null,"introduction.md","https:\u002F\u002Fgithub.com\u002Fteenygrad\u002Fteenygrad\u002Fedit\u002Fmain\u002Fbooks\u002Fteenygrad\u002Fsrc\u002Fintroduction.md","\u003Cblockquote>\n\u003Cp>⚠️ \u003Cstrong>Warning\u003C\u002Fstrong>: teenygrad is still under active development and not yet ready for production\nuse. This book, and the crates it documents, will change frequently.\u003C\u002Fp>\n\u003C\u002Fblockquote>\n\u003Cp>\u003Ca href=\"https:\u002F\u002Fteenygrad.org\" target=\"_blank\" rel=\"noopener noreferrer\">teenygrad\u003C\u002Fa> is a high-performance, memory-safe Rust ML training and\ninference library, in the spirit of PyTorch and tinygrad. It targets devices from\nmicrocontrollers to data centers with statically-typed GPU kernels and full async support.\u003C\u002Fp>\n\u003Ch2 id=\"vision\">Vision\u003C\u002Fh2>\n\u003Cp>We envision a world where machine learning is truly ubiquitous: from the tiniest embedded sensors\nto the largest distributed clusters, every device can harness the power of AI efficiently, safely,\nand without constraints.\u003C\u002Fp>\n\u003Cp>Goals:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>Highly concurrent, memory-safe\u003C\u002Fstrong> — designed for scalability and safety.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Low memory footprint\u003C\u002Fstrong> — optimized to run even on the smallest devices.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Statically typed\u003C\u002Fstrong> — GPU kernels and ML algorithms are statically typed.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>No performance compromises\u003C\u002Fstrong> — hardware-accelerated wherever possible.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Broad hardware support\u003C\u002Fstrong> — not just NVIDIA and AMD.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Extensible\u003C\u002Fstrong> — build high-performance accelerators for your own hardware.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Embedded-friendly\u003C\u002Fstrong> — core training\u002Finference components are \u003Ccode>no_std\u003C\u002Fcode> compatible\n(\u003Ca href=\"\u002Fapi\u002Fteenygrad\u002Fteenygrad\u002Fteeny_core\u002F\">\u003Ccode>teeny-core\u003C\u002Fcode>\u003C\u002Fa> builds \u003Ccode>no_std\u003C\u002Fcode> by default).\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Full async support\u003C\u002Fstrong>, multi-threaded by default.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2 id=\"how-this-book-is-organized\">How this book is organized\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>\u003Cstrong>Getting Started\u003C\u002Fstrong> walks through installing the toolchain and building your first model.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Core Concepts\u003C\u002Fstrong> covers the tensor\u002Fgraph types, the compilation pipeline, the dtype system, and\nname scoping — the ideas every other chapter builds on.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Neural Network Layers\u003C\u002Fstrong> covers the \u003Ccode>nn\u003C\u002Fcode> layer API.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Kernels & Backends\u003C\u002Fstrong> covers how kernels are written (the Triton-like DSL) and compiled to\ndevice code.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Compiler Internals\u003C\u002Fstrong> goes under the hood of the FXGraph compiler.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>PyTorch Interop\u003C\u002Fstrong> covers \u003Ccode>teeny-torch\u003C\u002Fcode>, the PyO3-based Python bridge.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>CLI & AOT Compilation\u003C\u002Fstrong> covers \u003Ccode>teeny-cli\u003C\u002Fcode> and \u003Ccode>teeny-llm\u003C\u002Fcode>.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Deployment\u003C\u002Fstrong> covers \u003Ccode>teeny-quant\u003C\u002Fcode>, weight-only model quantization for deployment.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>This book is a guide and mental model; it deliberately doesn’t duplicate API reference material —\nfor that, see the generated rustdoc at\n\u003Ca href=\"\u002Fapi\u002Fteenygrad\u002Fteenygrad\u002F\">docs.teenygrad.org\u002Fapi\u002Fteenygrad\u003C\u002Fa> (or \u003Ca href=\"https:\u002F\u002Fdocs.rs\" target=\"_blank\" rel=\"noopener noreferrer\">docs.rs\u003C\u002Fa>\nfor the crates that build there).\u003C\u002Fp>\n\u003Ch2 id=\"community\">Community\u003C\u002Fh2>\n\u003Cp>\u003Ca href=\"https:\u002F\u002Fdiscord.gg\u002FFG65XyPzuD\" target=\"_blank\" rel=\"noopener noreferrer\">Join our Discord\u003C\u002Fa> to discuss the project or get help.\u003C\u002Fp>\n",[12,16,19],{"id":13,"text":14,"level":15},"vision","Vision",2,{"id":17,"text":18,"level":15},"how-this-book-is-organized","How this book is organized",{"id":20,"text":21,"level":15},"community","Community",false,{"title":24,"titleHtml":25,"route":26},"Installation & Toolchain","Installation & Toolchain","\u002Fteenygrad\u002Fgetting-started\u002Finstallation",1786271828998]