Unverified50% confidenceFactExact time
NCM writes each model into a fixed-size output frame, so even if the answer is very short it must run through all layers and pay the full cost.
1
Sources
50%
Confidence
Long-term
Relevance
10/28/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedOutput limit refers to the generation space a model can use in one continuous task, while context window refers to how much material a model can read in; the two are different metrics.65% similarUnverifiedThe model used to run intelligent NPC dialogue on 4GB of VRAM is a quantized and compressed small-parameter model; the weights of a 3-billion-parameter model quantized to 4-bit can be compressed to under 2GB.64% similarUnverified考虑KV Cache、激活值和CUDA运行时开销后，7B模型总显存需求轻松突破20GB64% similarUnverifiedEvery time a model generates a token, the entire model must be read completely from memory, yielding the speed upper bound: memory bandwidth ÷ model size.62% similarUnverifiedFP16格式下，大模型显存需求的估算公式为：参数量 × 2 = 所需显存（GB），即每个参数以16位（2字节）浮点数存储62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1107923API
curl https://kongchang.com/api/v1/knowledge/claims/1107923MCP
get_claim(id=1107923)