Llm Inference Serving easy
A non-streaming endpoint generates reports of about 2,000 tokens at roughly 40 tokens/s per sequence. Your API gateway has a 30-second idle timeout, and many requests fail with 504 errors. What is the best fix?
- A Add more GPU replicas, since the extra hardware will make each individual report generate faster.
- B Increase the batch size, so the server finishes more reports per second and each completes sooner.
- C Stream tokens or use an async job, since a ~50 s generation cannot finish in a 30 s idle window.
- D Retry failed requests automatically, since the 504s are transient and a retry usually succeeds.
Answer: C. Stream tokens or use an async job, since a ~50 s generation cannot finish in a 30 s idle window.
At about 40 tokens/s, a 2,000-token report takes about 50 seconds, so a buffered response always exceeds a 30-second idle window. Streaming keeps the connection active, and an async job with polling decouples the work from the connection entirely. The tradeoff is more client complexity and a partial-output error model. Adding replicas is tempting but raises throughput, not single-sequence speed. Retries simply repeat the same 50-second generation and add load.
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?