How deeply does Domain-Specific Small Language Models cover quantization?

Explains precision formats, PTQ and QAT, schemes, hands-on absmax steps, 4-bit methods, and evaluation, without formal derivations.

Quantization

As explained in Domain-Specific Small Language Models

Methods that reduce numeric precision or representational size of model values to improve efficiency, especially for inference; here the absmax method is a specific quantization procedure.

Model quantization reduces the numeric precision of model weights or activations, typically replacing floating-point values with lower-bit representations. It trades some quantization error for lower memory use, potentially faster inference, and deployment on more constrained hardware. section 6.2 section 6.5

The main choices are precision format, quantization scheme, timing, and execution hardware. Post-training quantization (PTQ) modifies a trained model, while quantization-aware training (QAT) accounts for quantization during training; PTQ is especially practical for pretrained models that need only adaptation. section 6.1 section 6.2

Operational tradeoffs shape the deployment choice:

Use lower bit widths to reduce model memory requirements; 8-bit values can store roughly four times as many numbers as 32-bit floats in the same memory. section 6.2
Expect a balance among precision, memory footprint, inference speed, hardware support, and output quality rather than a universally best format. section 4.2 section 6.2
Choose domain-specific methods when ordinary quantization cannot satisfy the target environment; GPTQ supports efficient 3- or 4-bit weight quantization, while CPU and edge runtimes offer their own formats. section 6.4.1 section 11.4.2
Validate quality and serving behavior after conversion using measures such as perplexity, latency, throughput, and memory consumption. section 4.2 section 6.2.1

Absmax quantization

For an FP32 tensor X, calculate a scale as 127 divided by the tensor’s absolute maximum. Multiply X by that scale and round to obtain INT8 values; divide the rounded values by the scale to dequantize them. Applying this procedure layer by layer creates a dequantized model copy for comparison with the original. A GPT-2 example inspects weight histograms, generates text from both models, and compares perplexity; the quantized weights visibly differ, demonstrating that the method is lossy. section 6.2.1

Quantization fits into broader deployment pipelines. Models can be converted to ONNX, optimized or quantized, and then executed by a compatible runtime; specialized libraries such as AutoGPTQ, BitsAndBytes, and mllm target different model sizes and hardware environments. section 5.1 section 6.4.1 section 8.3

How Domain-Specific Small Language Models covers it

Cover of Domain-Specific Small Language Models

liveBook manning.com

Read Domain-Specific Small Language Models on liveBook Get the book at manning.com

Also explained in

liveBook manning.com

LLMs in ProductionQuantization is the conversion of model parameters or computations from higher-precision numerical representations to lower-precision representations.
liveBook manning.com

AI-Powered SearchQuantization is the transformation of numerical values into a smaller set of discrete values to reduce representation precision and resource requirements.
liveBook manning.com

Deep Learning with Python, Third EditionThe technique of reducing numeric precision of model weights or activations to speed up computation while preserving output behavior as much as possible.
liveBook manning.com

Deep Learning with R, Third EditionModel quantization is the conversion of model computations or parameters to lower-precision numerical representations to reduce resource requirements.
liveBook manning.com

Math and Architectures of Deep LearningThe conversion of floating-point region boundaries or ratios into integer-aligned indices for feature extraction.
liveBook manning.com

Generative AI in ActionQuantization is the reduction of numerical precision used to represent model parameters, typically to reduce memory use and computational cost.
liveBook manning.com

Deep Learning with PyTorch, Second EditionA model optimization technique that reduces numerical precision of parameters and operations to lower memory use and often speed up inference.
liveBook manning.com

Transformers in ActionQuantization is the transformation of numerical model parameters or activations into lower-precision representations to reduce memory consumption or computational cost.
liveBook manning.com

The Complete Obsolete Guide to Generative AIA quantized model is a machine-learning model whose numerical parameters or activations are represented at reduced precision to lower resource requirements.
liveBook manning.com

Time Series Forecasting Using Foundation ModelsQuantization is the conversion of continuous numerical values into a finite set of discrete levels or tokens.
liveBook manning.com

Natural Language Processing in Action, Second EditionQuantization is the transformation of numerical values into lower-precision discrete representations, often to reduce memory use and accelerate computation.
liveBook manning.com

Building Reliable AI SystemsQuantization reduces the numerical precision used to represent model values, often lowering storage or computation requirements.

Compare how 13 books explain quantization →

Related concepts

QLoRAused by · QLoRA is a parameter-efficient fine-tuning method that quantizes a pretrained model’s weights while training low-rank adapters to reduce memory requirements.
scalar quantizationincludes · Scalar quantization maps numerical scalar values to a finite set of discrete levels, typically reducing their precision or range.
QA-LoRAused by · QA-LoRA is a low-rank adaptation method that integrates quantization-aware operations and group-wise processing to support efficient low-bit adaptation.
post-training quantizationincludes · Post-training quantization is the conversion of a trained model’s parameters or computations to lower-precision representations without additional training.
quantization-aware trainingincludes · Quantization-aware training optimizes a model while simulating or incorporating reduced-precision effects so it adapts to quantized representations.
PyTorchuses · PyTorch is an open-source machine-learning framework for Python that provides tensor computation, automatic differentiation, and tools for building and training neural networks.

This explainer was generated by Manning's knowledge graph from the text of Domain-Specific Small Language Models. The section links point to the parts of the book it draws on.