MidnightTokensdeveloper portal
Sign In
Unit Study Document

Quantization Mechanics: GPTQ, AWQ & GGUF

8 min readβ€’Visual explainer included

Floating Point to Integer Quantization

Normally, weight values are stored as FP16 (2 bytes per parameter). A 7B parameter model requires 14GB of VRAM just to load. By quantizing parameters to 4-bit integers, we reduce VRAM to 3.5GB.

Fast Drill

Active Recalls

Card 1 of 1
Question

What is GGUF format best suited for?

Tap card to flip
Answer

CPU inference with partial GPU offloading, commonly used for running local LLMs on Macs/PCs.

Mastery: 0%
Knowledge Check

Quiz Practice

Question 1 of 1
Which format is optimized for GPU-only activation with structured weight lookup?

Chapter Scratchpad

Auto-saves immediately

Active Recall Cards

Review core concepts before doing the quiz

Fast Drill

Active Recalls

Card 1 of 1
Question

What is GGUF format best suited for?

Tap card to flip
Answer

CPU inference with partial GPU offloading, commonly used for running local LLMs on Macs/PCs.

Mastery: 0%

AI Study Buddy

Always online

Hi! I'm Spooky, your study buddy! Let's learn together.