Working on opt-in Windows HIP vision support for Strata. Image requests are working on my RX 7900 XTX with IQ3_XXS. Next: broader correctness testing and measuring the impact on text performance. Thanks @coldniko for building Strata!
Thanks to Strata I can now run two machines for Local AI instead of one. Also thankful to be discovering this in time for the cooler fall / winter months as summer would have been devastating for my home office.
Spent the day running glm5.3-flash on DS4 on the M5 128gb, and Qwen 3.8-flash-next IQ3_XSS w/Strata on the 7900 XTX. Impressed with the output of both.
I've primarily been running on an M5 128GB primarily Qwen 3.8-flash-next using mlx-serve, it's been very fast.
But checking out Strata on my AMD 7900 XTX , and this is running quite nice!
