Unverified50% confidenceFactExact time
以英语约5万常用词汇计算，Trigram 理论上需存储约1.25×10¹⁴种组合的概率
1
Sources
50%
Confidence
Long-term
Relevance
7/20/2026
First Seen
Sources
Related Claims
UnverifiedLLM生成阶段对词汇表（通常包含32,000至128,000个token）计算Softmax概率分布，形成总和为1的概率向量68% similarUnverifiedword2vec负采样中，对每个正样本随机采样k个（通常5-20个）负样本，采样概率通常按词频的3/4次方分布67% similarUnverifiedThe output of large models is probabilistic rather than deterministic; asking the same question 100 times may yield 100 different answers.66% similarUnverifiedA model with 100 billion parameters requires approximately 80 GB of VRAM to process a single inference request, while a model with tens of billions of parameters requires only about 20 GB.64% similarUnverifiedLarge model vocabularies typically range from 32,000 to 100,000 tokens.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/570278API
curl https://kongchang.com/api/v1/knowledge/claims/570278MCP
get_claim(id=570278)