Suggestion to Update PHI, Qwen, and BAAI Models for Improved Performance on Workers AI Platform

[anunol]

Currently, some of the models on the Workers AI platform, particularly in the PHI, Qwen, and BAAI series, are quite outdated and lagging in performance. Here are my specific suggestions:

1. PHI Series: I recommend replacing the PHI2 models on the Workers AI platform with the newer PHI3.5 series. The PHI3.5 models offer significantly improved performance and reasoning capabilities, and given that they have similar parameter sizes to PHI2, the cost increase would be minimal.

2. Qwen Series: The Qwen series has been updated to version 2.5, which significantly outperforms the 1.5 series in both efficiency and capability. I suggest deprecating the current Qwen1.5 models and replacing them with Qwen2.5. This will greatly enhance user experience and keep the platform aligned with cutting-edge technology from the same developer.

3. BAAI Series: I recommend upgrading the BGE1.5 series to the latest BAAI/bge-m3 models. The older models are increasingly inadequate in terms of performance, and replacing them will maintain high levels of user satisfaction on the platform.

I believe that replacing older models with the latest versions is a sensible approach, especially when the parameter sizes are comparable, leading to little or no increase in cost. However, the performance improvements are significant. The outdated models not only underperform but also discourage users from engaging with the Workers AI platform due to their suboptimal capabilities.

I hope to see these updates implemented soon. Thank you for your consideration!