medium10 pts · first answer countsmodel-servingperformance
GPU idle while CPU is maxed
A serving node shows GPU at 40% while one CPU core sits at 100%. Profiling shows time in tokenisation and detokenisation. What is a good fix?
Discussion 0
Join free to post in the discussion.
Related questions
- When batch inference fitsMLOps Engineer
- Traffic before the model has loadedMLOps Engineer
- Why export a model to ONNXMLOps Engineer
- Protocol for internal inference callsMLOps Engineer
- Latency and throughputMLOps Engineer
- What TTFT measuresMLOps Engineer
- Gap between streamed tokensMLOps Engineer
- Two phases of LLM generationMLOps Engineer