LESSON 8.4 · Systems · 160 min
DSpark: speculative generation and scheduling
DSpark: speculative generation and scheduling. Learn parallel drafting, semi-autoregressive heads, prefix survival, and hardware-aware scheduling and complete: Implement a prefix-s
Learning objectives
- Explain what problem “DSpark: speculative generation and scheduling” solves without hiding behind terminology.
- Trace the variables and causal links across parallel drafting, semi-autoregressive heads, prefix survival.
- Complete “Implement a prefix-survival scheduler under light and heavy load” and judge the result with evidence rather than intuition.
Core concepts
parallel drafting
parallel drafting 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.
semi-autoregressive heads
semi-autoregressive heads 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.
prefix survival
prefix survival 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 hardware-aware scheduling
and hardware-aware scheduling 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
Implement a prefix-survival scheduler under light and heavy load
- 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.