JavaScript seems to be disabled in your browser. For the best experience on our site, be sure to turn on Javascript in your browser.

Artículo: AMZ-B0G34QWNSG

Apple Silicon Intelligence: Mastering High-Performance Deep Learning, Unified Memory, and Lazy Evaluation with MLX.

Format:

Paperback

Hardcover

Kindle

Paperback

Detalles del producto

Disponibilidad
En stock

Peso con empaque
0.53 kg

Devolución
Sí

Condición
Nuevo

Producto de
Amazon

Viaja desde
USA

Sobre este producto

Unlock the Supercomputer on Your Desk. For the Past years, the "Iron Law" of deep learning was absolute: if you wanted to train serious models, you paid the "NVIDIA tax." You bought massive desktop towers, managed fragile CUDA drivers, or rented expensive cloud GPUs by the hour. Meanwhile, your MacBook Pro, a machine equipped with arguably the most advanced consumer silicon ever fabricated sat idle, relegated to checking emails and writing code for remote servers. The Iron Law has been broken. With the release of Apple’s M-series chips and the MLX framework, the paradigm has shifted. We are no longer bound by the PCIe bus bottleneck or the separation of CPU and GPU memory. We have entered the era of Unified Memory Architecture, where a single pool of high-bandwidth RAM is accessible to the CPU, the GPU, and the Neural Engine simultaneously. What You Will Learn: This book takes you far beyond the "Hello World" of MNIST. It deconstructs the MLX framework from the first principles of hardware design to the bleeding edge of Large Language Model (LLM) fine-tuning.The Hardware Reality: You will stop viewing your Mac as a consumer laptop and start treating it as a NUMA-like workstation. You will learn why memory bandwidth (400GB/s+) matters more than clock speed, how the M3 Max compares to the RTX 4090 in real-world transformer workloads, and how to exploit the "Zero-Copy" architecture to eliminate data transfer latency entirely.Thinking in MLX: You will unlearn the imperative habits of PyTorch and master the functional, lazy evaluation model of MLX. You will learn how to visualize dynamic computation graphs, how to use mx.compile to fuse Python logic into blazing-fast Metal kernels, and how to manage stateless randomness for perfectly reproducible research.High-Performance Mathematics: We dive deep into the implementation of functional transformations. You will learn to replace slow Python loops with mx.vmap (vectorization), implement custom automatic differentiation with mx.grad, and write your own custom GPU operations using the Metal Shading Language (MSL) and C++.Engineering LLMs from Scratch: The core of the book focuses on the modern Transformer architecture. You will build, optimize, and run massive language models locally.Flash Attention: Implement memory-efficient attention that allows for massive context windows without crashing your RAM.The KV Cache: Master the state management required for high-speed text generation.RoPE Embeddings: Understand and implement Rotary Positional Embeddings to handle long-context sequences.Fine-Tuning 70B Models: You will learn the specific techniques required to fine-tune state-of-the-art models (like Llama-3-70B) on a single machine. We cover the full implementation of LoRA (Low-Rank Adaptation) and QLoRA, teaching you how to mix 4-bit quantized base weights with high-precision adapters to democratize model training.Forensic Debugging: When things go wrong, you need more than print statements. You will learn to use the Metal Debugger in Xcode to visualize GPU command buffers, profile memory usage to detect retention cycles, and optimize "Time To First Token" (TTFT) latency down to the millisecond.Who This Book Is For: This book is written for the engineer who wants to own the stack.You are a Machine Learning Engineer who wants to prototype and train locally without latency.You are a Researcher who needs to implement novel architectures that don't fit the rigid constraints of standard libraries.You are a Software Developer building local-first AI applications that need to run efficiently on user hardware without relying on APIs.Stop waiting for the cloud. Build it locally.

$76,45

55% OFF

$34,75

$76,45

55% OFF

$34,75

Envío gratis

Llega en 5 a 12 días hábiles

Con envío

Tienes garantia de entrega

Medios de pago

Cantidad

Este producto viaja de USA

a tus manos en

Medios de pago

Tarjetas de Débito, Crédito y Deuna