Paraphrase MiniLM L6 V2
Neural Network
Paraphrase MiniLM L6 v2 is an embedding model that converts sentences and short paragraphs into 384-dimensional vectors
Specifications
- Input
- text
- Output
- vectors (embeddings)
- Context size
- 512 tokens
- Vector dimensions
- 384
Prompt cost
(per 1M tokens)
How it works Paraphrase MiniLM L6 V2?
Paraphrase MiniLM L6 v2 encodes sentences and short paragraphs into dense 384-dimensional vectors, optimized for paraphrase search, semantic similarity assessment, clustering, and lightweight database search. Its compact size saves vector storage memory and speeds up retrieval. In BotHub, the model works via a unified OpenAI-compatible API, without VPN or foreign cards, with pay-as-you-go billing in rubles. Switching to another embedder from the catalog does not require rewriting the integration. Connect it to Qdrant or pgvector for RAG, build semantic search for documentation and tickets, cluster reviews, and find duplicates or semantically similar entries in your knowledge base.Frequently asked questions about Paraphrase MiniLM L6 V2
You can use generated results for commercial purposes. You own all rights to the content you create. The only restriction: make sure your prompt does not include copyrighted third-party material. You are responsible for respecting the rights to any input data.
This is an embedding model from sentence-transformers: it converts text into a numerical vector used to measure the semantic similarity of fragments. It is suitable for semantic search, RAG-based knowledge base search, ticket clustering, deduplication, and paraphrase search. It does not perform text generation.
The output is not text, but an array of numbers of fixed dimensionality — a compact vector that is easy to store and index in Qdrant, pgvector, FAISS, or other vector storage. The input window is 512 tokens, so split long documents into chunks beforehand.
Step-by-step reasoning mode is not indicated for this model in our data, and it is not needed for its task: the query is encoded into a vector in a single pass. If you need a chain of thought, connect a reasoning text model in the same BotHub window.
Function calling and structured output are not indicated for this model in our data. However, the API response comes in a standard JSON structure with an array of embeddings, making it easy to integrate the model into a pipeline: it uses the unified OpenAI-compatible BotHub endpoint.
Acceptance of modalities other than text is not indicated in our data — vectorize text fragments. If your project needs images or audio, specialized models are available in BotHub, and you can switch to them without rewriting the integration, using the same API key.
We do not publish data on the language coverage of this specific version, so you should check the quality on a Russian-language corpus yourself: run your sample and compare search metrics with another embedding model. In BotHub, you pay per token, without a subscription.