LLM推理系统问答
How would you debug a sudden TTFT p99 spike in a production LLM serving system?
原题What is goodput and why is it more meaningful than raw throughput or individual metric SLO compliance?
原题How would you design an A/B testing framework for comparing two LLM versions in production?
原题Explain how you would capacity plan for an LLM serving system expecting 10× user growth in 6 months.
原题How would you measure and improve Model FLOP Utilization (MFU)?
原题How do diffusion language models generate text differently from autoregressive models? What are the inference implications?
原题What is Mixture of Depths? How does it enable adaptive inference compute?
原题Why might photonic computing change LLM inference fundamentally? What are the current limitations?
原题How does multi-token prediction improve both training and inference efficiency?
原题Describe an “inference-first” model design philosophy. What architectural choices reflect this approach?
原题
暂无题目