Amazon: Titan Embeddings G1 - Text

{"about":"### Text meaning as a numerical vector\n\nAmazon Titan Embeddings G1 is Amazon’s text embedding model, also known as V1. It converts text into a vector of **1,536 numbers**. This lets an application compare phrases and documents by meaning even when they use different words—for example, finding instructions for updating account details in response to a question about changing payment information.\n\n### Search, knowledge bases and clustering\n\nG1 can prepare a text index and create a search representation of a user’s query. In a knowledge-base system, this is one stage of RAG: relevant chunks are retrieved first, then a separate language model uses them to compose an answer. Titan itself returns a vector at this stage; document storage, similarity calculations and answer construction belong to your application.\n\nAnother use case is text clustering. Vectors can help find similar support requests, organize material by topic or identify related documents. Specific categories and similarity thresholds are defined by the workflow; an embedding does not contain ready-made labels.\n\n### How to prepare your material\n\n1. Divide the collection into meaningful chunks: sections, paragraphs or self-contained answers.\n2. Obtain vectors for those chunks and retain a link between each vector and its source text.\n3. Convert the search query in the same way.\n4. Compare it with the index and return the retrieved material to the user or the next processing stage.\n\nAiloot API provides text input of up to **8,192 tokens** and embedding output. For search, choose chunk size to suit the content: a single large vector for an entire document can obscure differences between the individual questions it covers. When the source text changes, update its representation in the index as well.\n\n### What distinguishes G1\n\nThis generation of Titan uses a fixed 1,536-dimensional representation. Amazon describes training on 25 natural languages, with the most extensive testing in English. For specialist terminology, ambiguous wording and other languages, use a small set of your own queries with known relevant documents. G1 is primarily intended for natural-language text; for source-code search, the next Titan generation has separate information about training on programming languages.","alias_of":null,"aliases":[],"author":"Amazon","canonical_model":"amazon/titan-embed-text-v1","ctx":8192,"formats":["embeddings"],"id":"amazon/titan-embed-text-v1","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":0.2,"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-v1","route_label":"Ultra","status_reads":[],"tiers":[],"tool_prices":null,"unit":"token","units":[]}],"model_aliases":[{"canonical_model_id":"amazon/titan-embed-text-v1","request_model_id":"amazon/titan-embed-text-v1","route_label":"Ultra"}],"name":"Amazon: Titan Embeddings G1 - Text","price":{"in":0.2,"in_label":null,"out":null,"out_label":null,"sort_in":0.2,"sort_out":null,"varies":false},"public_id":"amazon/titan-embed-text-v1","short":"Converts text into 1,536-dimensional vectors for semantic search, finding similar material and clustering documents."}