GPTQ and AWQ Post-Training Quantization Quiz

Answers and explanations appear as you take the quiz.

1. What does 'post-training quantization' mean?
   • Compressing an already-trained model's weights to lower precision without full retraining
   • Retraining the model entirely at low precision
   • Adding more training data after deployment
   • Increasing the model's parameter count
2. What information does GPTQ use to compensate for rounding errors when quantizing?
   • The model's vocabulary list
   • Second-order (Hessian/curvature) information from a calibration dataset
   • The user's prompt history
   • Random noise
3. What key insight drives AWQ (Activation-aware Weight Quantization)?
   • All weights are equally important
   • Activations should be quantized but weights left untouched
   • A small fraction of weight channels are especially important and should be protected
   • Quantization should happen during pretraining only
4. Roughly how much memory does 4-bit quantization save versus 16-bit weights?
   • No savings
   • About double the memory
   • Exactly half the memory
   • About one quarter the memory (a 4x reduction)
5. Why do both GPTQ and AWQ typically quantize weights but keep activations at higher precision?
   • Weights dominate memory, and aggressive activation quantization tends to hurt accuracy more
   • Activations don't exist in transformers
   • Activations are already 1-bit
   • It's required by law
6. Which tool is commonly used to run quantized models locally on consumer hardware?
   • Microsoft Word
   • llama.cpp
   • A spreadsheet program
   • A web browser alone
7. How does GPTQ use a calibration dataset during layer-wise weight quantization?
   • It adds new training labels to the model
   • It selects a new tokenizer vocabulary
   • It estimates second-order information to compensate for rounding errors
   • It measures only GPU temperature
8. How does AWQ identify which weight channels deserve protection?
   • It computes a full Hessian inverse
   • It adds random noise to all weights
   • It selects channels only by their index
   • It uses activation magnitudes to identify channels important to preserve