AI Systems · core
Model Quantization
Post-training and quantization-aware methods, integer and low-bit formats, calibration, kernels, and quality trade-offs.
Mental model
Quantization stores and computes approximate weights or activations with fewer bits. The serving win is real only when hardware kernels support the format and evals bound quality loss.
How to study Model Quantization
Begin by restating the mental model in your own words, then connect it to a concrete system you have built or operated. Name the mechanism, the constraint it addresses, and the trade-off it introduces. Use AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration, LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale, SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models to check details, but close the source before writing your explanation. Retrieval is the learning step; rereading is only preparation.
Next, compare Model Quantization with LoRA & PEFT, Multimodal Models. Ask what changes in correctness, latency, resource use, operability, and failure recovery. Complete Design exercise: Model Quantization and preserve the command, input, output, and one failed attempt as evidence. Finish by explaining the idea without jargon to someone who has not studied the track.
Proof of understanding
- Explain the mechanism from first principles and identify the state it reads or changes.
- Give one situation where the concept is the right choice and one where it is not.
- Predict a realistic failure mode before running the drill, then compare the prediction with evidence.
- Connect the result to a roadmap or build artifact instead of treating the concept as isolated trivia.
Learn from primary sources
- AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration (doc)
- LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale (doc)
- SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models (doc)
- Democratizing Foundation Models via k-bit Quantization — Tim Dettmers (Stanford MLSys #82) (video)
- GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers (paper)
- Hugging Face Transformers — Quantization (doc)
Practice and explain it back
Design exercise: Model Quantization
Post-training and quantization-aware methods, integer and low-bit formats, calibration, kernels, and quality trade-offs. Implement designOutline() returning non-empty values for: numericFormat, calibration, qualityGate. Each value must name a concrete mechanism or decision.
Expected evidence: A design outline with numericFormat, calibration, qualityGate plus an explicit failure mode or trade-off.
Open the interactive drill →Review prompts
- You quantized to int4 and memory dropped but throughput did not. What went wrong?
Build evidence
Synthesize: AI Models & Training
Move from transformer foundations through pre-training, fine-tuning, post-training, compression, and evaluation. Produce one working system, benchmark, or evidence-backed design that integrates the path.
- Implements or precisely models the core mechanisms from all three milestones
- Includes at least one injected failure or adversarial case and demonstrates recovery
- Reports quality, latency, resource, reliability, or usability measurements relevant to the domain
- Ships a concise architecture note explaining decisions, trade-offs, and remaining risks
Synthesize: Inference & Serving
Build a production mental model for inference engines, memory, kernels, routing, hardware, and serving economics. Produce one working system, benchmark, or evidence-backed design that integrates the path.
- Implements or precisely models the core mechanisms from all three milestones
- Includes at least one injected failure or adversarial case and demonstrates recovery
- Reports quality, latency, resource, reliability, or usability measurements relevant to the domain
- Ships a concise architecture note explaining decisions, trade-offs, and remaining risks
Synthesize: Multimodal & Spatial Computing
Connect vision, audio, generation, on-device intelligence, robotics, spatial interfaces, and HCI. Produce one working system, benchmark, or evidence-backed design that integrates the path.
- Implements or precisely models the core mechanisms from all three milestones
- Includes at least one injected failure or adversarial case and demonstrates recovery
- Reports quality, latency, resource, reliability, or usability measurements relevant to the domain
- Ships a concise architecture note explaining decisions, trade-offs, and remaining risks