Issues / #2149

#2149 [Bug] v0.1.43 release broke Strata on AMD 7000 series card. Weights loaded to VRAM get offloaded to RAM after a few seconds.

open · @PARLTOR · 0 comments · View on GitHub

Setup & installAMD / HIPModels & quants

Description

v0.1.43 release broke Strata on AMD 7000 series card. Weights loaded to VRAM get offloaded to RAM after a few seconds. Leading RAM usage to spike until limit (64Gb in my case) => blue screen, sluggish, etc.

Clean re-install of latest release v0.1.43 => same behavior across IQ3_xxs and IQ3_s. Across the different context length options and KV quantization choices.

7900 XTX 24Gb on Win11

Was working well and v0.1.41 and previous

Related on strata.com

Editorial links to help you install, pick models, or read release notes — not part of the upstream thread.