-
Notifications
You must be signed in to change notification settings - Fork 2
Description
Hi, I'm trying to get CUDA support for my open source image scanning app. And it's proving a nightmare. :) My dev setup is: Intel Core i7-8700, 64GB RAM, RTX5060ti, Python 3.12, latest CUDA, and obviously 16GB VRAM. The problem is I've tried all sorts of prebuilt .whl options and none work. The latest attempt generated an app error of:
OSError: [WinError -1073741795] Windows Error 0xc000001d
ERROR:services.analysis_service:Job 3d95901b-98c5-4c46-b051-b11f7c4edc10 failed: Failed to load model medgemma-27b-it-Q4_K_S
Opus 4.5 says this is a problem with my CPU only supporting AVX2 and not AVX-512. So I've been trying - and pitifully failing - to build something around compatible specs. It seems that Opus can't get it together at all. Sigh.
Any help would be very much appreciated.