Llm Inference Serving medium
Analytics show 30% of chat users click 'stop' or close the tab mid-response. Your SSE server keeps generating each abandoned response until max_tokens. GPU cost is rising and queue wait is growing. What should you change?
- A Detect the closed connection and abort the sequence in the scheduler, freeing its batch slot and KV blocks.
- B Lower max_tokens for all users so that abandoned responses consume fewer decode steps before they finish.
- C Buffer abandoned outputs server-side so that users who return can be served the finished response instantly.
- D Add an idle timeout so the load balancer drops streams that have had no client reads for 60 seconds.
Answer: A. Detect the closed connection and abort the sequence in the scheduler, freeing its batch slot and KV blocks.
**Propagating cancellation** to the scheduler reclaims the KV memory and batch slot immediately. The cost is plumbing disconnect signals through proxies and deciding how to bill partial output. A global max_tokens cut (B) truncates legitimate long answers, and a load-balancer timeout (D) drops the socket while generation keeps running unless the cancel reaches the engine. Finishing generation (C) is right only for async APIs where the client polls for a stored result.
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?