Alibaba's Qwen2-7B-based GTE that topped MTEB English and Chinese in mid-2024.
Access model weights, configuration files, and documentation.
See which devices can run this model and at what quality level.
The General Text Embedding (GTE) model from Alibaba's Tongyi Lab, built atop Qwen2-7B with bidirectional attention and query-side instruction tuning. Trained via multi-stage contrastive learning on a large multilingual corpus, it ranked #1 on both MTEB English and C-MTEB on its June 2024 release; produces 3584-dim embeddings and supports up to 32K context.
View on Hugging Face