Sentence Transformers: all-mpnet-base-v2

Sentence Transformers: all-mpnet-base-v2 is an embedding model from Sentence Transformers, a family of models built for turning text into vectors. It sits in the embeddings category, meaning it is designed to produce representations of text rather than to generate text itself. The model encodes sentences and short paragraphs into a 768-dimensional dense vector space, and it is positioned by its provider as a high-fidelity option for semantic work.

On the input side, the model accepts text only. On the output side, it returns embeddings rather than words or chat messages, so you should not expect it to answer questions or hold a conversation. The context length is 512 tokens, and the maximum output is 460. Both numbers matter in practice: the 512 token window covers a sentence or a short paragraph, and anything longer has to be split before you send it. The embedding it returns is a 768-dimensional vector.

Those capabilities point to a fairly specific set of uses. Information retrieval is a natural fit, where you embed a query and a corpus and compare them. Clustering is another, since sentences that sit close together in the 768-dimensional space often belong to the same topic. Similarity scoring follows directly from that, for example ranking candidate passages against a reference, or detecting near-duplicate text. Because the model handles sentences and short paragraphs, these tasks work best when each input you embed is short and self-contained.

You call it through api14's OpenAI-compatible API, using the model id "sentence-transformers/all-mpnet-base-v2". Because the interface follows the OpenAI embeddings shape, existing clients and libraries that target that endpoint can be pointed at api14 without special handling. Note that this model lists no supported parameters, so the request body should carry the model id and the input text, and you should not rely on optional knobs such as dimensions or encoding format being applied.

The 512 token context and the 460 output cap are the main practical limits here. If your text is longer than a sentence or a short paragraph, split it into chunks and embed each one, then combine the results. And if what you actually need is generated text rather than vectors, a generative chat model will fit better; this model only produces embeddings.