# The Metrale Engine Book > Documentation for Metrale Engine — an open source, pure Rust and CUDA LLM inference > engine. One ~75 MB binary, no Python, no PyTorch. This is the reference documentation. It covers installing and running the engine, the architecture of the kernel dispatch pipeline, per-crate references, and operational guides. The rustdoc API reference is generated from the source and published alongside it. ## Part I — Getting Started - [Philosophy](https://book.dev.metrale.ai/getting-started/philosophy.html) - [Installation](https://book.dev.metrale.ai/getting-started/installation.html) - [Quickstart](https://book.dev.metrale.ai/getting-started/quickstart.html) - [Supported Models](https://book.dev.metrale.ai/getting-started/models.html) - [Troubleshooting](https://book.dev.metrale.ai/getting-started/troubleshooting.html) ## Part II — Architecture - [Philosophy: AI Kernel HyperCompiling](https://book.dev.metrale.ai/architecture/philosophy.html) - [Workspace Layout](https://book.dev.metrale.ai/architecture/workspace.html) - [Kernel Dispatch Pipeline](https://book.dev.metrale.ai/architecture/dispatch.html) - [The Circuit Compiler](https://book.dev.metrale.ai/architecture/circuit-compiler.html) - [The Latent Kernel Blueprint](https://book.dev.metrale.ai/architecture/lkb.html) - [The Latent Architecture Blueprint](https://book.dev.metrale.ai/architecture/lab.html) - [SBIO: Business Logic vs I/O](https://book.dev.metrale.ai/architecture/sbio.html) ## Part III — The Crates - [metrale-core](https://book.dev.metrale.ai/crates/metrale-core.html) - [metrale-config](https://book.dev.metrale.ai/crates/metrale-config.html) - [metrale-closure](https://book.dev.metrale.ai/crates/metrale-closure.html) - [metrale-governance](https://book.dev.metrale.ai/crates/metrale-governance.html) - [metrale-kernels](https://book.dev.metrale.ai/crates/metrale-kernels.html) - [metrale-gpu-sys](https://book.dev.metrale.ai/crates/metrale-gpu-sys.html) - [metrale-telemetry](https://book.dev.metrale.ai/crates/metrale-telemetry.html) - [metrale-gpu-runtime](https://book.dev.metrale.ai/crates/metrale-gpu-runtime.html) - [metrale-scheduler](https://book.dev.metrale.ai/crates/metrale-scheduler.html) - [metrale-cache](https://book.dev.metrale.ai/crates/metrale-cache.html) - [metrale-sampling](https://book.dev.metrale.ai/crates/metrale-sampling.html) - [metrale-storage](https://book.dev.metrale.ai/crates/metrale-storage.html) - [metrale-comm](https://book.dev.metrale.ai/crates/metrale-comm.html) - [metrale-grammar](https://book.dev.metrale.ai/crates/metrale-grammar.html) - [metrale-model-weights](https://book.dev.metrale.ai/crates/metrale-model-weights.html) - [metrale-model-layers](https://book.dev.metrale.ai/crates/metrale-model-layers.html) - [metrale-model-arch](https://book.dev.metrale.ai/crates/metrale-model-arch.html) - [metrale-model-engine](https://book.dev.metrale.ai/crates/metrale-model-engine.html) - [metrale-speculative](https://book.dev.metrale.ai/crates/metrale-speculative.html) - [metrale-bench](https://book.dev.metrale.ai/crates/metrale-bench.html) - [metrale-server](https://book.dev.metrale.ai/crates/metrale-server.html) ## Part IV — Engineering Deep Dives - [CUDA Kernel Engineering](https://book.dev.metrale.ai/deep-dives/kernels.html) - [NVFP4 Quantization](https://book.dev.metrale.ai/deep-dives/nvfp4.html) - [FP8 Native Serving](https://book.dev.metrale.ai/deep-dives/fp8.html) - [Attention & Paged KV Cache](https://book.dev.metrale.ai/deep-dives/attention.html) - [MoE Routing & Experts](https://book.dev.metrale.ai/deep-dives/moe.html) - [SSM / Mamba / GDN Layers](https://book.dev.metrale.ai/deep-dives/ssm.html) - [Speculative Decoding (MTP)](https://book.dev.metrale.ai/deep-dives/mtp.html) - [Constrained Decoding (XGrammar)](https://book.dev.metrale.ai/deep-dives/xgrammar.html) ## Part V — Operating Metrale Engine - [OpenAI-Compatible Server](https://book.dev.metrale.ai/operations/server.html) - [Tool Calling & Streaming](https://book.dev.metrale.ai/operations/tools.html) - [Multi-GPU & EP=2](https://book.dev.metrale.ai/operations/multi-gpu.html) - [Benchmarking](https://book.dev.metrale.ai/operations/benchmarks.html) - [Certification](https://book.dev.metrale.ai/operations/certify.html) ## Part VI — The Project - [Contributing](https://book.dev.metrale.ai/project/contributing.html) - [How a Change Lands](https://book.dev.metrale.ai/project/landing.html) - [The Merge Lattice](https://book.dev.metrale.ai/project/merge-lattice.html) - [Security Policy](https://book.dev.metrale.ai/project/security.html) - [Release Notes](https://book.dev.metrale.ai/project/releases.html) ## Appendix - [A Category-Theoretic Perspective](https://book.dev.metrale.ai/appendix/category-theory.html) - [The LKB in Mathematics](https://book.dev.metrale.ai/appendix/lkb-math.html) - [Glossary](https://book.dev.metrale.ai/appendix/glossary.html) - [Further Reading](https://book.dev.metrale.ai/appendix/reading.html) ## Machine-readable - [Full API reference](https://docs.metrale.ai/api/): rustdoc for every public crate - [Print view](https://book.dev.metrale.ai/print.html): the entire book as one HTML document ## Optional - [Metrale Engine](https://metrale.ai/engine): the project page — also at https://metrale.ai/llms.txt - [Engineering blog](https://blog.metrale.ai): measured notes — also at https://blog.metrale.ai/llms.txt - [Source](https://github.com/Metrale/metrale-inference): pure Rust and CUDA, MIT OR Apache-2.0 - [Recipes](https://github.com/Metrale/metrale-inference/tree/main/recipes): the model SSOT - [Discord](https://discord.gg/RQcGakU2jW)