hard20 pts · first answer countsmodel-servingparallelismmixture-of-experts
Placing experts on many GPUs
A large mixture-of-experts model has 256 experts per layer, too many for one GPU. With expert parallelism, what is the main communication cost?
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