Appendix D · Glossary [DRAFT]¶
[To write: Glossary.]
Planned contents¶
Latency terms: TTFT, ITL, TPOT, e2e latency, goodput
Engine terms: prefill, decode, step, continuous batching, iteration-level scheduling, preemption
Memory terms: KV cache, block, block table, paged attention, fragmentation, copy-on-write
Optimisation terms: GQA/MQA/MLA, quantisation, calibration, speculation, acceptance rate
Scaling terms: tensor/pipeline/expert parallelism, disaggregation, prefix affinity