Unverified50% confidenceFactExact time
For VLMs, the visual encoding stage concentrates heavy computation during the prefill phase, while the language decoding stage is limited by memory bandwidth.
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
Unverified文本生成阶段是memory-bound问题，速度受限于内存带宽；预填充阶段是compute-bound场景，可高度并行化75% similarVerified大语言模型推理分预填充和解码两阶段：预填充是计算密集型对GPU算力敏感，解码是内存带宽受限任务73% similarVerifiedThe Prefill phase is a compute-intensive operation, while the Decode phase is a memory-bandwidth-intensive operation.72% similarUnverifiedAt the hardware level, the Prefill (input) stage can read all tokens in parallel and compute the self-attention matrix simultaneously, whereas the Decode (output) stage requires serial autoregressive token-by-token inference, causing GPU utilization to drop sharply.72% similarUnverifiedLLM inference requests are divided into two phases—Prefill (input) and Decode (output); Prefill can read inputs in parallel with extremely high throughput, while Decode requires token-by-token serial inference and occupies significantly more GPU time.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/902378API
curl https://kongchang.com/api/v1/knowledge/claims/902378MCP
get_claim(id=902378)