Llm Inference Serving easy
Your API lets clients omit max_tokens, and the default is the full 8K-token context. Most replies are under 300 tokens, but a few runaway generations loop for thousands of tokens. What does setting a sensible default cap, such as 1K tokens, accomplish?
- A It mainly speeds up prefill, since the server sizes the prompt's KV cache based on the max_tokens value.
- B It bounds worst-case decode time and KV memory per request, at the risk of cutting off long valid answers.
- C It has no effect on capacity, since KV cache is allocated only for tokens the model actually generated.
- D It lowers inter-token latency for every request, since each decode step is cheaper with a smaller cap.
Answer: B. It bounds worst-case decode time and KV memory per request, at the risk of cutting off long valid answers.
A cap bounds the **worst case**: runaway generations can no longer tie up a decode slot and thousands of tokens of KV cache, both of which other requests need. The tradeoff is truncating legitimately long outputs, so long-form use cases should set a higher explicit limit. The 'no effect' answer is tempting because paged allocation is on demand, but runaways really do consume that memory, and some schedulers reserve KV based on max_tokens.
One system design question a day, with spaced repetition so the ones you miss come back.
Get the appRelated questions
- Your 70B-class model servers each handle about 40 concurrent chats with 4K-token contexts. Product raises the maximum context to 32K tokens, and many users start using it. What is the main capacity consequence?
- A 70B-class model (80 layers, grouped-query attention with 8 KV heads of dimension 128) is served in fp16 on one 8-GPU node with about 640 GB of total HBM. Product wants 256 concurrent sessions, each holding a full 32K-token context. What is the binding constraint, and what does it imply?
- Your server reserves a contiguous KV region sized for max_tokens = 8,192 for every request at admission, but the average sequence ends near 600 tokens. The node reports out-of-KV-memory at about 20 concurrent requests while GPU compute sits at 30%. Which change best raises concurrency without restricting legitimate long outputs?
- You must summarize 10 million documents overnight, and no user is waiting on any single result. How should you configure the serving stack compared with your interactive chat deployment?
- You serve a 7B-parameter model in fp16 (about 14 GB of weights) for a chat app. Users say single-conversation token generation is too slow, and you are choosing new GPU hardware. Which hardware spec most directly increases per-sequence decode speed?
- Your RAG workload sends 8K-token prompts and gets ~150-token answers from a 70B-class model. You plan to move prefill and decode onto separate GPU pools so long prefills stop stalling decode. What new resource must you size carefully?