jina-embeddings-v2-small-en supports plain English text as input, typically provided as strings representing sentences, paragraphs, or longer text segments. In practice, developers pass UTF-8 encoded text without special markup requirements. The model does not require structured formats like JSON or XML for the text itself; instead, those formats are usually part of the surrounding API or application logic, not the model input. As long as the content is English text, the model can process it.
In real applications, input text often comes from sources such as documentation pages, support tickets, emails, product descriptions, or chat transcripts. Developers usually preprocess this text by cleaning up excessive whitespace, removing irrelevant boilerplate, and splitting long documents into smaller chunks. Chunking is important because embedding extremely long text can dilute semantic focus and exceed token limits. Once chunked, each piece of text is embedded independently and stored as a separate vector, often in a system like Milvus or Zilliz Cloud, along with metadata such as document ID or section title.
The model does not natively understand formatting like Markdown or HTML tags in a semantic way, so it is generally recommended to strip or normalize markup before embedding. For example, converting HTML to plain text or removing Markdown syntax can improve consistency. The key point for developers is that jina-embeddings-v2-small-en expects clean, readable English text and will treat whatever it receives as raw content. Keeping preprocessing simple and consistent usually leads to better retrieval quality when those embeddings are later queried in a vector database.
For more information, click here： https://zilliz.com/ai-models/jina-embeddings-v2-small-en