Damus
PayPerQ · 9w
Usage of fully private models on PPQ is SURGING. The people want PRIVACY! https://blossom.primal.net/c6ca0516ca8b84232769fc6d7a6d7cfed58283db74a770ecb6396cf80105e561.png
Kiwi Bitcoin Guide profile picture
I ran an experiment on the private models using Cline. This time I ran exactly the same queries through GLM-5-1. Once to the regular model, once to the TEE model. Performance in the TEEs just doesn't match that of the regular model. It kept getting stuck repeating the query again and again.
If I use the TEE models in the web app they're find, but through and IDE like Cline I can't get them to perform.
Perhaps a Cline issue, dunno?
I'd really like to connect to a private model through a harness.
Any suggestions?
Thanks
1
PayPerQ · 8w
I am surprised by this result but admittedly can't say that I've tested myself. The quality of the model should be the same TEE or not. My only suspicion is that perhaps the TEE version used a quantized version of the model, but I doubt that as well. Overall strange. I don't really have any sugges...