Unverified50% confidenceFactExact time
A text token roughly corresponds to 0.75 English words or approximately 1.5 Chinese characters, and image visual encoders typically split a single image into hundreds of visual tokens.
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
Unverified在Stable Diffusion中，CLIP文本编码器将自然语言切分为最多77个token，每个token映射为768维或1024维嵌入向量68% similarUnverified上下文窗口决定模型单次能处理的文本量：8K约6000词，32K约24000词，128K可处理整本书内容64% similarVerifiedtext-embedding-ada-002将文本转化为1536维的稠密向量，语义相近的文本在向量空间中距离更近63% similarUnverifiedImage-to-Text leverages the visual understanding capabilities of multimodal large language models to map visual features to natural language descriptions.62% similarUnverified令牌是大语言模型处理文本的基本计量单位，英文中一个令牌约对应0.75个单词62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/902376API
curl https://kongchang.com/api/v1/knowledge/claims/902376MCP
get_claim(id=902376)