Unverified50% confidenceFactExact time
The system uses a dual-index design in which documents are split into 600-character chunks with 100-character overlap, with NOMIC embed text generating vector embeddings stored in ChromaDB.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/7/2026
First Seen
Valid until: 1/5/2027
Sources
Related Entities
Related Claims
Unverified常见的文档分片方式有四种：按字数划分、按段落划分、按章节划分、按页码划分64% similarUnverified现代LLM的分词器通常将文本压缩为约10万词汇量的Token序列61% 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61% similarUnverifiedlangchain-text-splitters is responsible for splitting long documents into text chunks suitable for embedding models and large language model context windows.60% similarUnverifiedCLIP文本编码器的标准上下文窗口长度约为77个Token，超出部分的语义信息会被截断或降权处理60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/983697API
curl https://kongchang.com/api/v1/knowledge/claims/983697MCP
get_claim(id=983697)