Alibaba unveiled the newest artificial intelligence (AI) embedding model ‘QWEN3 Embedding Series’ on the sixth (local time) as an open source.
The embedding model is a core technology that converts text into numbers to assist computers understand meaning. It enables more precise and meaningful search results.
The released model supports greater than 100 languages, including multilingual and programming languages, and features powerful multilingual processing, cross language search and code search ability.
Particularly, it was ranked first within the Massive Text Embedding Benchmark, which evaluates text embedding performance.
The Q1 3 -embedding series is anticipated to reinforce the educational optimization of the prevailing Q1 model and the efficiency of the embedding and economic system system. The ‘Reranking’ function can be strengthened to make the search results more precisely to user queries.
The three -stage learning structure applied to the prevailing Alibaba model was followed. The primary stage relies on the vast raw data, and the contrast learning process is assessed based on the relevance, and within the second stage, the refined high -quality data verifies the system’s judgment ability. Finally, these two processes are integrated to maximise the performance of the model.
Alibaba defined the Kyu -One 3 Impeding series as a “latest start line,” and said, “We look ahead to applying this model to more developers.”
By Park Chan, reporter cpark@aitimes.com
