Amazon: Titan Text Embeddings V2

{"about":"### The second generation of Titan text embeddings\n\nAmazon Titan Text Embeddings V2 converts text into a numerical representation for search and other tasks based on semantic similarity. The model is optimized for retrieving relevant chunks: a user asks a question in their own words, and the application compares its vector with the vectors of knowledge-base material. This makes it possible to find information without an exact keyword match.\n\nV2 differs from G1 in its more compact default representation: Amazon’s documentation specifies a default vector size of **1,024 dimensions**. For an application, this is a property of the data it stores and compares. Every entry in an index must use a consistent representation; when switching from another embedding model, rebuild the index for the selected model.\n\n### Where to use it\n\n- **RAG and enterprise search:** create vectors for documentation sections, then retrieve chunks for a separate model that answers the question.\n- **Semantic similarity:** compare differently worded texts and find related material.\n- **Clustering:** explore topics in support requests, reviews and documents through the similarity of their representations.\n- **Code search:** build search workflows for source code and its accompanying explanations.\n\nAmazon states that V2 was trained on more than 100 natural languages and 13 programming languages, including Python, JavaScript, Java, C++, Rust, Go and SQL. The model is nevertheless optimized primarily for English text. In a multilingual knowledge base, specifically test queries where the question and documents are in different languages: including a language in training does not establish equal quality across every such combination.\n\n### Working with Ailoot API\n\nThe input is text and the output is an **embedding**. This listing has a limit of **8,192 input tokens**. For a search index, divide long material at meaningful boundaries such as headings, sections and paragraphs. Each chunk can then be stored alongside its vector and a reference to its original source.\n\n### A practical workflow\n\nSelect real questions and relevant answers from your collection, create embeddings and check the search results. Adjust chunk size and the comparison method using the same evaluation set, while retaining the link to the source text. Titan produces numerical representations; comparison, result selection and answer generation take place in subsequent application stages. This evaluation helps you choose index settings suited to your terminology and document structure.","alias_of":null,"aliases":[],"author":"Amazon","canonical_model":"amazon/titan-embed-text-v2","ctx":8192,"formats":["embeddings"],"id":"amazon/titan-embed-text-v2","input":["text"],"kind":"embedding","max_out":0,"methods":[{"cache_read":null,"cache_write":null,"cache_write_1h":null,"cache_write_5m":null,"conditional_meter":null,"ctx":8192,"family":"openai","fast_multiplier":1,"flex_multiplier":1,"fmt":"embeddings","http":"POST","img_in":null,"img_out":null,"in":0.04,"input":["text"],"label":"Эмбеддинги","max_out":0,"max_tool_calls":null,"out":0,"output":["embedding"],"path":"/v1/embeddings","request_model_id":"amazon/titan-embed-text-v2","route_label":"Ultra","status_reads":[],"tiers":[],"tool_prices":null,"unit":"token","units":[]}],"model_aliases":[{"canonical_model_id":"amazon/titan-embed-text-v2","request_model_id":"amazon/titan-embed-text-v2","route_label":"Ultra"}],"name":"Amazon: Titan Text Embeddings V2","price":{"in":0.04,"in_label":null,"out":null,"out_label":null,"sort_in":0.04,"sort_out":null,"varies":false},"public_id":"amazon/titan-embed-text-v2","short":"Creates text vectors for RAG, semantic search and clustering, including search across multilingual content and code."}