LESSON 8.5 · Theory · 130 min
Diffusion and block-parallel language models
Diffusion and block-parallel language models. Learn masked diffusion, parallel decoding, block diffusion, and quality-speed tradeoffs and complete: Compare latency models for autor
Learning objectives
- Explain what problem “Diffusion and block-parallel language models” solves without hiding behind terminology.
- Trace the variables and causal links across masked diffusion, parallel decoding, block diffusion.
- Complete “Compare latency models for autoregressive, parallel, and diffusion drafting” and judge the result with evidence rather than intuition.
Core concepts
masked diffusion
masked diffusion is part of the lesson’s causal model. State its inputs, outputs, invariants, and failure mode; then verify it with a hand-check or a minimal experiment before moving to an optimized implementation.
parallel decoding
parallel decoding is part of the lesson’s causal model. State its inputs, outputs, invariants, and failure mode; then verify it with a hand-check or a minimal experiment.
block diffusion
block diffusion is part of the lesson’s causal model. State its inputs, outputs, invariants, and failure mode; then verify it with a hand-check or a minimal experiment.
and quality-speed tradeoffs
and quality-speed tradeoffs is part of the lesson’s causal model. State its inputs, outputs, invariants, and failure mode; then verify it with a hand-check or a minimal experiment.
Build and verify
Compare latency models for autoregressive, parallel, and diffusion drafting
- Predict: write the expected output, trend, or failure before running code.
- Build: implement only the minimum components needed to answer the question.
- Verify: compare with a baseline or trusted implementation; save seeds, parameters, and raw outputs.
- Transfer: change one shape, dataset, scale, or workload condition and explain whether the conclusion still holds.