Publication Specifications
FORMATDRM-Free PDF
LENGTH35 Pages
TIERadvanced Edition
PUBLISHERBlizarra Ltd
Advanced35 pages • Instant Download
Local LLMs on Apple Silicon & Linux
Running Ollama, llama.cpp, vLLM, and quantized weights privately with maximum throughput.
One-time purchase
£39.00
Instant Access
Encrypted payment processing via Stripe Checkout
DRM-free PDF e-book delivered immediately after payment
Purchase receipt emailed automatically
About this Book
Total data sovereignty and zero cloud bills. Master the art of local inference on Mac M-series Unified Memory and Linux GPU servers. Understand GGUF vs EXL2 quantization, KV-cache quantization, continuous batching, and integrating local endpoints into standard OpenAI-compatible SDKs.
Target Audience: Privacy-focused engineers, researchers, and systems programmers who want local inference.
What You Will Learn & Build
Harness Apple Unified Memory bandwidth for 32B and 70B parameter models
Configure llama.cpp compile flags for Metal, CUDA, and ROCm hardware acceleration
Deploy vLLM with PagedAttention and continuous batching for production concurrency
Expose local models via private OpenAI-compatible proxy servers with authentication
Table of Contents
1. Hardware Fundamentals: Memory Bandwidth vs Compute FLOPSSection 1
2. Quantization Formats: GGUF, AWQ, GPTQ and EXL2 DemystifiedSection 2
3. llama.cpp Deep Dive: Compilation, Context Sizes & KV OffloadingSection 3
4. Ollama for Rapid Prototyping & Docker ContainerizationSection 4
5. High-Concurrency Serving with vLLM on Linux GPUsSection 5
6. Secure Local Network Access & Team Reverse ProxiesSection 6
7. Fine-Tuning & LoRA Adapters on Consumer HardwareSection 7
8. Building a Local Multi-Model RouterSection 8