Alibaba Ai Researchers Released A New Gte-qwen2-7b-instruct Embedding Model Based On The Qwen2-7b Model With Better Performance

Alibaba researchers have released a new embedding model, gte-Qwen1.5-7B-instruct, which is based on Qwen2-7B and has shown improved performance on the Massive Text Embedding Benchmark (MTEB). This benchmark covers 8 embedding tasks, 58 datasets, and 112 languages, and no single TE method has been found to work best for all tasks. The new model has an overall score of 70.24 and nDCG@10 for Retrieval of 60.25 on the MTEB leaderboard.