Inquiry about Future Plans for GTE-Qwen Models Based on Qwen2.5
Hello,
First, thank you for providing such a powerful and versatile model in the GTE-Qwen series. I am currently considering using this model for my academic research. However, I noticed that while GTE-Qwen2-7B-Instruct has demonstrated top-tier performance in the MTEB benchmark, it appears that other models may now surpass it in the current MTEB ranking.
Given the advancements introduced in the Qwen2.5 series, such as improved instruction-following, extended multilingual support, and long-context processing (up to 128k tokens), I believe there is significant potential for a GTE model based on Qwen2.5 to outperform the current version.
Additionally, if such an updated GTE model could accept larger token sequences (up to 128k) and potentially increase the embedding dimensions, it would greatly enhance its utility for complex, large-scale research tasks.
Are there any plans to release an updated GTE-Qwen model based on Qwen2.5? If so, I would be very interested in learning more about its potential features and timeline.
Thank you for your time and for continuing to support the community with these cutting-edge models!