# Text Embedding

Gemini Embedding is an experimental text embedding model launched by Google, provided through the Gemini API. This model demonstrates outstanding performance in the Multilingual Text Embedding Benchmark (MTEB), surpassing previous top models. It can convert text into high-dimensional numerical vectors, capturing semantic and contextual information, and is widely used in scenarios such as retrieval, classification, and similarity detection. Gemini Embedding supports over 100 languages, features an 8K input token length and 3K output dimension, and incorporates Multi-Representation Learning (MRL) technology, allowing for flexible dimension adjustment to meet storage requirements. The model is currently in the experimental stage, and a stable version will be released in the future.

Trieve Vector Inference is an on-premises vector inference solution designed to address the issues of high latency and high failure rates in text embedding services. It allows users to host dedicated embedding servers in their own cloud, enabling faster text embedding inference. This product helps enterprises reduce their dependency on external services and improve data processing speed and efficiency by providing high-performance on-premises inference services.

E5-mistral-7b-instruct is a text embedding model with 32 layers and an embedding size of 4096. It can be used to encode queries and documents into semantic vector representations. The model uses natural language task descriptions to guide the text embedding process, allowing for customization based on different tasks. It is trained on the MS-MARCO passage ranking dataset and can be applied to information retrieval, question answering, and other natural language processing tasks.

Jina Embeddings V2 Base is an English text embedding model that supports a sequence length of 8192. It is based on the Bert architecture (JinaBert) and supports the ALiBi symmetric bidirectional variant to allow for longer sequence lengths. The model was pre-trained on the C4 dataset and further trained on a collection of over 400 million sentence pairs and negative samples from Jina AI. This model is suitable for various use cases involving long documents, including long document retrieval, semantic text similarity, text re-ranking, recommendation, RAG, and LLM-based generative search. The model has 137 million parameters and is recommended for inference on a single GPU.