Unverified50% confidenceSolutionExact time
PD disaggregation deploys the compute-intensive Prefill stage and the memory-bandwidth-intensive Decode stage of inference on separate instances for independent scaling and optimization
1
Sources
50%
Confidence
Long-term
Relevance
10/4/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedLLM 推理分为预填充阶段（Prefill，计算密集型）和解码阶段（Decode，内存带宽密集型）两个阶段75% similarUnverifiedDCUF (Decompose, Classify, Unite, Filter) adopts a decomposition-based strategy that breaks claims into sub-questions, retrieves evidence for each separately, and then merges the results.67% similarUnverifiedThe prefill stage is compute-bound, requiring all input tokens to be processed at once to generate the KV Cache, with attention matrix computation complexity of O(n²).66% similarUnverifiedThe Prefill stage of LLM inference is a compute-intensive operation whose primary bottleneck is GPU/NPU compute (FLOPS).66% similarUnverifiedThe prefill phase of large language model inference is computationally intensive and can fully utilize a large number of parallel cores.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/972990API
curl https://kongchang.com/api/v1/knowledge/claims/972990MCP
get_claim(id=972990)