Jinaai
Models by this creator
🔎
jina-embeddings-v2-base-en
625
The jina-embeddings-v2-base-en model is a text embedding model created by Jina AI. It is based on a BERT architecture called JinaBERT that supports longer sequence length up to 8192 tokens using the symmetric bidirectional variant of ALiBi. The model was further trained on over 400 million sentence pairs and hard negatives from various domains. This makes it useful for a range of use cases like long document retrieval, semantic textual similarity, text reranking, and more. Compared to the smaller jina-embeddings-v2-small-en model, this base version has 137 million parameters, allowing for fast inference while delivering better performance. Model inputs and outputs Inputs Text sequences up to 8192 tokens long Outputs 4096-dimensional text embeddings Capabilities The jina-embeddings-v2-base-en model can generate high-quality embeddings for long text sequences, enabling applications like semantic search, text similarity, and document understanding. Its ability to handle 8192 token sequences makes it particularly useful for working with long-form content like research papers, legal contracts, or product descriptions. What can I use it for? The embeddings produced by this model can be used in a variety of downstream natural language processing tasks. Some potential use cases include: Long document retrieval: Finding relevant documents from a large corpus based on semantic similarity to a query. Semantic textual similarity: Measuring the semantic similarity between text pairs, which can be useful for applications like plagiarism detection or textual entailment. Text reranking: Reordering a list of documents or passages based on their relevance to a given query. Recommendation systems: Suggesting relevant content to users based on the semantic similarity of items. RAG and LLM-based generative search: Enabling more powerful and flexible search experiences powered by large language models. Things to try One interesting aspect of the jina-embeddings-v2-base-en model is its ability to handle very long text sequences, up to 8192 tokens. This makes it well-suited for working with long-form content like research papers, legal contracts, or product descriptions. You could try using the model to perform semantic search or text similarity analysis on a corpus of long-form documents, and see how the performance compares to models with shorter sequence lengths. Another interesting area to explore would be the model's use in recommendation systems or generative search applications. The high-quality embeddings produced by the model could be leveraged to suggest relevant content to users or to enable more flexible and powerful search experiences powered by large language models.
Updated 5/27/2024
🌐
reader-lm-1.5b
251
reader-lm-1.5b is a series of models developed by Jina AI that convert HTML content to Markdown content. The models are trained on a curated collection of HTML content and its corresponding Markdown content, allowing them to effectively perform content conversion tasks. There are two main models in the reader-lm series: reader-lm-0.5b with a context length of 256K reader-lm-1.5b with a context length of 256K These models can be used to convert HTML content to Markdown format, which is useful for tasks like content migration, blog post formatting, and more. Model inputs and outputs Inputs HTML content: The model takes raw HTML content as input, with no prefix instruction required. Outputs Markdown content: The model outputs the corresponding Markdown version of the input HTML content. Capabilities The reader-lm models are capable of effectively converting HTML content to Markdown format, leveraging their training on a curated dataset of HTML-Markdown pairs. This allows them to accurately preserve the structure and formatting of the original HTML content when generating the Markdown output. What can I use it for? The reader-lm models can be a valuable tool for a variety of content-related tasks, such as: Content migration**: Easily convert HTML content to Markdown format when moving content between platforms or websites. Blog post formatting**: Automatically convert HTML blog posts to Markdown, which is a common format for many blogging and publishing platforms. Document conversion**: Convert HTML documentation or reports to Markdown for better readability and portability. Things to try One interesting thing to try with the reader-lm models is to explore their performance on different types of HTML content, such as complex web pages, long-form articles, or even code-heavy documentation. You can also experiment with the models' ability to preserve formatting, links, and other HTML elements when generating the Markdown output.
Updated 9/18/2024
🖼️
jina-clip-v1
165
jina-clip-v1 is a state-of-the-art English multimodal (text-image) embedding model trained by Jina AI. It bridges the gap between traditional text embedding models, such as jina-embeddings-v2-base-en, which excel in text-to-text retrieval but are incapable of cross-modal tasks, and models like openai/clip-vit-base-patch32 that effectively align image and text embeddings but are not optimized for text-to-text retrieval. jina-clip-v1 offers robust performance in both domains, matching the retrieval efficiency of jina-embeddings-v2-base-en for text-to-text tasks while setting a new benchmark for cross-modal retrieval. Model inputs and outputs Inputs Sentences**: The model can encode meaningful sentences in English. Images**: The model can also encode images, either by providing the public image URLs or directly passing in the PIL.Image objects. Outputs Text embeddings**: The model outputs dense vector representations for the input sentences. Image embeddings**: The model outputs dense vector representations for the input images. Similarity scores**: The model can compute the cosine similarity between text and image embeddings, enabling cross-modal retrieval. Capabilities jina-clip-v1 excels at both text-to-text and text-to-image retrieval tasks. Its dual capability makes it an excellent tool for multimodal retrieval-augmented generation (MuRAG) applications, allowing seamless text-to-text and text-to-image searches within a single model. What can I use it for? jina-clip-v1 can be used for a variety of multimodal applications, such as: Image search**: Users can search for images by describing them in text. Cross-modal retrieval**: The model can retrieve relevant text or images based on a query in the opposite modality. Multimodal question answering**: The model can be used to answer questions that require understanding both text and images. Multimodal content generation**: The model can be used to generate relevant text or images based on a prompt in the opposite modality. Jina AI has also provided the Embeddings API as an easy-to-use interface for working with jina-clip-v1 and their other embedding models. Things to try One key advantage of jina-clip-v1 is its ability to handle longer sequences of text, up to 8,192 tokens, thanks to its use of the symmetric bidirectional variant of ALiBi. This makes the model well-suited for tasks involving long-form content, such as document retrieval, long-form question answering, and summarization. Researchers and developers can explore how the model's performance scales with longer input sequences compared to traditional text embedding models.
Updated 7/2/2024
👨🏫
jina-reranker-v2-base-multilingual
133
The jina-reranker-v2-base-multilingual model is a transformer-based text reranking model trained by Jina AI. It is a cross-encoder model that takes a query and a document pair as input and outputs a score indicating the relevance of the document to the query. The model is trained on a large dataset of query-document pairs and is capable of reranking documents in multiple languages with high accuracy. Compared to the previous jina-reranker-v1-base-en model, the Jina Reranker v2 has demonstrated competitiveness across a series of benchmarks targeting text retrieval, multilingual capability, function-calling-aware and text-to-SQL-aware reranking, and code retrieval tasks. Model inputs and outputs Inputs Query**: The input query for which relevant documents need to be ranked Documents**: A list of documents to be ranked by relevance to the input query Outputs Relevance scores**: A list of scores indicating the relevance of each document to the input query Capabilities The jina-reranker-v2-base-multilingual model is capable of handling long texts with a context length of up to 1024 tokens, enabling the processing of extensive inputs. It also utilizes a flash attention mechanism to improve the model's performance. What can I use it for? You can use the jina-reranker-v2-base-multilingual model for a variety of text retrieval and ranking tasks, such as improving the search experience in your applications, enhancing the performance of your information retrieval systems, or integrating it into your AI-powered decision support systems. The model's multilingual capability makes it a suitable choice for global or diverse user bases. Things to try To get started with the jina-reranker-v2-base-multilingual model, you can try using the Jina AI Reranker API. This provides a convenient way to leverage the model's capabilities without having to worry about the underlying implementation details. You can also explore integrating the model into your own applications or experimenting with fine-tuning the model on your specific data and use case.
Updated 7/31/2024
💬
jina-embeddings-v2-base-zh
121
The jina-embeddings-v2-base-zh model is a Chinese/English bilingual text embedding model developed by Jina AI. It is based on a BERT architecture (JinaBERT) that supports the symmetric bidirectional variant of ALiBi to allow longer sequence lengths of up to 8192 tokens. Compared to other Jina embedding models, jina-embeddings-v2-base-zh is a 161 million parameter model trained specifically on mixed Chinese-English input to provide high performance in both mono-lingual and cross-lingual applications. Similar Jina AI embedding models include jina-embeddings-v2-small-en, jina-embeddings-v2-base-en, jina-embeddings-v2-base-de, and an upcoming jina-embeddings-v2-base-es model for Spanish-English bilingual embeddings. Model Inputs and Outputs Inputs Text sequence**: The model takes in text sequences of up to 8192 tokens, supporting both Chinese and English, as well as a mix of the two. Outputs Text embeddings**: The model outputs 768-dimensional embedding vectors that capture the semantic meaning of the input text. These can be used for a variety of downstream tasks like information retrieval, text similarity, and multilingual applications. Capabilities The jina-embeddings-v2-base-zh model has been designed to excel at both mono-lingual and cross-lingual tasks involving Chinese and English text. Its long sequence length support of up to 8192 tokens makes it useful for applications that need to process long-form content, such as document retrieval, semantic textual similarity, and text reranking. What Can I Use It For? The jina-embeddings-v2-base-zh model can be used for a wide range of natural language processing tasks that require high-quality text embeddings, especially those involving a mix of Chinese and English text. Some potential use cases include: Information Retrieval**: Use the embeddings for semantic search and retrieval of Chinese or English documents, or documents containing a mix of both languages. Text Similarity**: Compute the similarity between Chinese, English, or bilingual text passages to detect paraphrases, identify related content, or perform clustering. Multilingual Applications**: Leverage the model's cross-lingual capabilities to build applications that seamlessly handle Chinese and English input, such as chatbots or question-answering systems. Things to Try An interesting aspect of the jina-embeddings-v2-base-zh model is its ability to handle long input sequences of up to 8192 tokens. This makes it well-suited for tasks involving lengthy documents or multi-paragraph inputs. You could experiment with using the model for tasks like: Long-form text summarization, where the model's ability to capture semantic meaning in long passages could improve the quality of generated summaries. Cross-lingual document retrieval, where the model's bilingual capabilities and long sequence support could help surface relevant content even when the query and target documents are in different languages. Multilingual dialog systems, where the model's embeddings could be used to maintain context and coherence across language switches within a conversation. By exploring the model's unique features, you can uncover novel applications that leverage its strengths in handling long, multilingual text inputs.
Updated 5/28/2024
🔎
jina-embeddings-v2-small-en
110
jina-embeddings-v2-small-en is an English text embedding model trained by Jina AI. It is based on a BERT architecture called JinaBERT that supports longer sequence lengths of up to 8192 tokens using the ALiBi technique. The model was further trained on over 400 million sentence pairs and hard negatives from various domains. Compared to the larger jina-embeddings-v2-base-en model, this smaller 33 million parameter version enables fast and efficient inference while still delivering impressive performance. Model inputs and outputs Inputs Text sequences**: The model can handle text inputs up to 8192 tokens in length. Outputs Sentence embeddings**: The model outputs 768-dimensional dense vector representations that capture the semantic meaning of the input text. Capabilities jina-embeddings-v2-small-en is a highly capable text encoding model that can be used for a variety of natural language processing tasks. Its ability to handle long input sequences makes it particularly useful for applications like long document retrieval, semantic textual similarity, text reranking, recommendation, and generative search. What can I use it for? The jina-embeddings-v2-small-en model can be used for a wide range of applications, including: Information Retrieval**: Encoding long documents or queries into semantic vectors for efficient similarity-based search and ranking. Recommendation Systems**: Generating embeddings of items (e.g. articles, products) or user queries to enable content-based recommendation. Text Classification**: Using the sentence embeddings as input features for downstream classification tasks. Semantic Similarity**: Computing the semantic similarity between text pairs, such as for paraphrase detection or question answering. Natural Language Generation**: Incorporating the model into RAG (Retrieval-Augmented Generation) or other LLM-based systems to improve the coherence and relevance of generated text. Things to try A key advantage of the jina-embeddings-v2-small-en model is its ability to handle long input sequences. This makes it well-suited for tasks involving lengthy documents, such as legal contracts, research papers, or product manuals. You could explore using this model to build intelligent search or recommendation systems that can effectively process and understand these types of complex, information-rich text inputs. Additionally, the model's strong performance on semantic similarity tasks suggests it could be useful for building chatbots or dialogue systems that need to understand the meaning behind user queries and provide relevant, context-aware responses.
Updated 5/28/2024
🧠
reader-lm-0.5b
82
The reader-lm-0.5b model is a series of models from Jina AI that can convert HTML content to Markdown content. This is useful for content conversion tasks, as the model is trained on a curated collection of HTML and corresponding Markdown content. The model is available in two sizes: reader-lm-0.5b and reader-lm-1.5b, which have 256K context lengths. The similar models in this series include the reader-lm-1.5b model, which has the same context length as the reader-lm-0.5b model. Both models can be loaded and used in a similar way. Model inputs and outputs Inputs Raw HTML content Outputs Markdown content corresponding to the input HTML Capabilities The reader-lm-0.5b model can convert HTML content to Markdown format, which is useful for tasks such as content migration, formatting, and processing. The model can handle a wide range of HTML structures and produce clean, well-formatted Markdown output. What can I use it for? The reader-lm-0.5b model can be used in a variety of content conversion and processing tasks. For example, you could use it to convert blog posts, articles, or other web content from HTML to Markdown format, making it easier to work with the content in a variety of tools and platforms. The model could also be used as part of a content management system or web scraping pipeline to automatically convert HTML content to a more portable format. Things to try One interesting thing to try with the reader-lm-0.5b model is to experiment with the input HTML content and see how the model handles different types of structures and formatting. You could try feeding the model a range of HTML content, from simple pages to more complex, nested structures, and observe how the Markdown output varies. This could help you understand the model's capabilities and limitations, and identify any areas for improvement or fine-tuning. Another thing to try is to use the model as part of a larger content processing pipeline, integrating it with other tools and services to create a more comprehensive content management workflow. For example, you could use the model to convert HTML to Markdown, and then use the Markdown content as input to a text summarization or natural language processing model to extract key insights or generate related content.
Updated 9/18/2024
🎲
jina-colbert-v1-en
76
Jina-ColBERT Jina-ColBERT is a variant of the ColBERT retrieval model that is based on the JinaBERT architecture. Like the original ColBERT, Jina-ColBERT uses a late interaction approach to achieve fast and accurate retrieval. The key difference is that Jina-ColBERT supports a longer context length of up to 8,192 tokens, enabled by the JinaBERT backbone which incorporates the symmetric bidirectional variant of ALiBi. Model inputs and outputs Inputs Text passages to be indexed and searched Outputs Ranked lists of the most relevant passages for a given query Capabilities Jina-ColBERT is designed for efficient and effective passage retrieval, outperforming standard BERT-based models. Its ability to handle long documents up to 8,192 tokens makes it well-suited for tasks involving large amounts of text, such as document search and question-answering over long-form content. What can I use it for? Jina-ColBERT can be used to power a wide range of search and retrieval applications, including enterprise search, academic literature search, and question-answering systems. Its performance characteristics make it particularly useful in scenarios where users need to search large document collections quickly and accurately. Things to try One interesting aspect of Jina-ColBERT is its ability to leverage the JinaBERT architecture to support longer input sequences. Practitioners could experiment with using Jina-ColBERT to search through long-form content like books, legal documents, or research papers, and compare its performance to other retrieval models.
Updated 5/28/2024
📶
jina-colbert-v2
70
The jina-colbert-v2 model is a new version of the JinaColBERT retrieval model developed by Jina AI. It builds upon the capabilities of the previous jina-colbert-v1-en model by adding multilingual support, improved efficiency and performance, and new Matryoshka embeddings that allow flexible trade-offs between precision and efficiency. Like its predecessor, jina-colbert-v2 uses a token-level late interaction approach to achieve high-quality retrieval results. The model is an upgrade from the English-only jina-colbert-v1-en, with expanded support for dozens of languages while maintaining strong performance on major global languages. It also includes the improved efficiency, performance, and explainability benefits of the JinaBERT architecture and ALiBi that were introduced in the previous version. Model inputs and outputs Inputs Text to be encoded, up to 8192 tokens in length Outputs Contextual token-level embeddings, with options for 128, 96, or 64 dimensions Ranking scores for retrieval, leveraging the late interaction mechanism Capabilities The jina-colbert-v2 model offers superior retrieval performance compared to the jina-colbert-v1-en model, particularly for longer documents. Its multilingual capabilities and flexible embeddings make it a versatile tool for a variety of neural search applications, including long-form document retrieval, semantic search, and question answering. What can I use it for? The jina-colbert-v2 model can be used to power neural search systems that require high-quality retrieval from large text corpora, including use cases like: Enterprise search**: Indexing and retrieving relevant documents from an organization's knowledge base E-commerce search**: Improving product and content discovery on online marketplaces Question answering**: Retrieving the most relevant passages to answer user queries The model's support for long input sequences and multiple languages makes it particularly well-suited for handling complex, multilingual search tasks. Things to try Some key things to explore with the jina-colbert-v2 model include: Evaluating the different embedding sizes**: The model offers 128, 96, and 64-dimensional embeddings, allowing you to experiment with the trade-off between precision and efficiency. Leveraging the Matryoshka embeddings**: The model's Matryoshka embeddings enable flexible retrieval, where you can balance between precision and speed as needed. Integrating the model into a broader neural search pipeline**: The jina-colbert-v2 model can be used in conjunction with other components like rerankers and language models to create a end-to-end neural search system.
Updated 9/18/2024
🚀
jina-embeddings-v2-base-code
57
The jina-embeddings-v2-base-code model is a multilingual text embedding model trained by Jina AI. It supports 8192 sequence length and can encode text in English and 30 widely used programming languages. The model is based on a BERT architecture (JinaBERT) that uses the symmetric bidirectional variant of ALiBi to allow for longer sequence length. The backbone jina-bert-v2-base-code model was pretrained on the github-code dataset. It was then further trained on Jina AI's collection of over 150 million coding question-answer and docstring source code pairs from various domains. This allows the model to effectively encode code and technical language. Jina AI also provides several other embedding models that differ in size and language support, including jina-embeddings-v2-small-en, jina-embeddings-v2-base-en, jina-embeddings-v2-base-zh, jina-embeddings-v2-base-de, and jina-embeddings-v2-base-es. Model inputs and outputs Inputs Text**: The model can take English or programming language text as input, with a maximum sequence length of 8192 tokens. Outputs Text embeddings**: The model outputs 384-dimensional dense vector representations that capture the semantic meaning of the input text. These embeddings can be used for a variety of downstream tasks like semantic search, text ranking, and question answering. Capabilities The jina-embeddings-v2-base-code model is particularly well-suited for applications that require encoding long-form technical text, such as programming code, documentation, and scientific literature. Its ability to handle sequences up to 8192 tokens makes it useful for tasks that involve processing entire documents or passages, rather than just short sentences or paragraphs. What can I use it for? The jina-embeddings-v2-base-code model can be used for a variety of applications, including: Code search and retrieval**: The model can be used to encode code snippets or docstrings and perform semantic search to find relevant code examples. Technical question answering**: The model's ability to encode long-form technical text can be leveraged to build systems that can answer complex questions by retrieving and understanding relevant information from a knowledge base. Documentation summarization**: The model's embeddings can be used to identify the most salient parts of long technical documents, enabling efficient summarization. Recommendation systems**: The model's embeddings can be used to find related programming resources, such as libraries, frameworks, or online tutorials, based on a user's interests or the context of their current project. Things to try One interesting aspect of the jina-embeddings-v2-base-code model is its ability to handle longer sequences of text, up to 8192 tokens. This makes it well-suited for tasks that involve processing entire documents or code repositories, rather than just individual sentences or paragraphs. For example, you could try using the model to encode a large codebase and then perform semantic search to find relevant code snippets for a specific programming task. Or you could use the model to generate embeddings for technical papers or blog posts, and then use those embeddings to power a recommendation system that suggests related content to users. Another potential application is using the model's embeddings as input to a language model or question-answering system, to build a more powerful technical assistant that can understand and respond to complex queries about programming, software engineering, or scientific topics.
Updated 9/18/2024