By Sandeep Belgavi · Learning path · Updated 28 October 2026
Begin with memory arithmetic, experiment with numerical error, then distinguish adapters from the frozen base model.
Recommended sequence
- How quantization changes the memory budget
Beginner · no prerequisites. Evidence: Explanation + calculation.
- Why loading a model does not prove a request fits
Beginner · weight-memory basics. Evidence: Explanation + cache-budget lab.
- Compare four-bit and eight-bit numerical error
Intermediate · basic rounding. Evidence: Explanation + synthetic vector lab.
- Inspect what LoRA and QLoRA train
Intermediate · matrix dimensions. Evidence: Explanation + parameter-count lab.
- Understand the serving architecture
Intermediate · inference-memory basics. Evidence: Technical explanation.
The interactive calculations run in the article pages. They use synthetic inputs and do not contact a model. Measured GPU results are added only after successful recorded runs.
How to interpret evidence
An explanation develops a concept using references and reasoning. A simulation shows the output of a simplified model with explicit assumptions. A measured experiment reports an actual run, its configuration, sample size and limitations. A simulation does not establish production performance. Follow each guide’s prerequisites and check the stated expected outcome before changing an input.
All new examples use generic technical tasks and synthetic data. No account, personal dataset or business scenario is required for these browser calculations. Source references remain available in the guides. Choose another learning path.