Llm Inference Serving medium
Your 64K-context document Q&A service runs a model whose 8-bit weights take 20 GB per GPU, which leaves 55 GB for an fp16 KV cache. KV memory caps concurrency at a few sequences per GPU. Which change most increases concurrent sequences per GPU?
- A Quantize weights further to 4-bit, since weights are the largest fixed memory cost and freeing them makes room for KV.
- B Quantize the KV cache to 8-bit, roughly doubling the tokens that fit, after evaluating quality on long-context tasks.
- C Shrink the maximum batch size so each running sequence gets a larger share of the GPU's KV memory pool.
- D Add a second replica on another GPU, which doubles the KV memory available to each long-context sequence.
Answer: B. Quantize the KV cache to 8-bit, roughly doubling the tokens that fit, after evaluating quality on long-context tasks.
At 64K context the **KV cache dominates memory**. 8-bit KV halves the bytes per token and roughly doubles capacity, at the cost of some attention precision that should be checked on long-context evaluations. 4-bit weights (A) free only about 10 GB, roughly 18% more KV, while adding more risk to quality. That lever wins when contexts are short and weights are the bulk of memory.
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?