Quiz with answers — KV cache, recurrent memory, MLA, and low rank
Answer all eight questions, then check the score. Open only the explanations needed for remediation.
1. Solve the worked-case variant: Double context length from 4,096 to 8,192 tokens while keeping 32 layers, 8 KV heads, d_head=128, and BF16. Recompute the simplified raw cache.
Show answer
B. Cache grows from 536,870,912 bytes (512 MiB) to 1,073,741,824 bytes (1 GiB). Growth is linear in token count; MLA or recurrent state can reduce different dimensions but also changes fidelity or retrieval behavior. The correct answer executes the requested change and gives a checkable result; the other texts do not close this calculation or trace.
2. Which causal order correctly connects the first three stages of “KV cache, recurrent memory, MLA, and low rank”?
Show answer
C. Standard KV cache → Fixed recurrent state → MLA The chain follows the taught progression; reversing stages consumes a representation or state before it is produced.
3. If “MLA” is removed, which diagnostic method is defensible?
Show answer
A. Keep the same input, predict the first output that depends on “MLA,” then compare the before/after trace. One intervention and a prior prediction make the delta attributable to the removed mechanism.
4. Which verdict respects this session’s validity boundary?
Show answer
D. Simplified formulas omit alignment, quantization, buffers, and sharing details. They compare trends, not promise real footprint. The correct answer bounds the conclusion; the others turn a local relation into a global guarantee.
5. Which evidence best matches the stated status of “KV cache, recurrent memory, MLA, and low rank”?
Show answer
B. Mixed: established mechanisms + source-reported Kimi K3-style choices. Product or mechanism evidence must remain attributed and measured; availability and completion do not prove value.
6. When should a simpler baseline be preferred to “MLA is not fixed memory”?
Show answer
C. When a controlled test shows equivalent quality with lower memory, latency, or complexity. The choice depends on a measured trade-off on the real workload, not novelty or one isolated metric.
7. A learner gets the right result but cannot explain “Fixed recurrent state.” Which remediation is most useful?
Show answer
A. Rebuild the first missing transformation, label its inputs and outputs, then test a neighboring case. The remediation targets the first causal break and then requires transfer instead of rewarding a guessed result.
8. Which submission actually demonstrates the outcome “Distinguish architectural low rank from LoRA.”?
Show answer
D. A trace with starting data, transformations, observed result, boundary, and next experiment. The correct submission makes the reasoning reproducible and the verdict revisable by future measurement.