From cdcc083f7fb5fa983793758cac9f13967165e87e Mon Sep 17 00:00:00 2001 From: Ean Garvey <87458719+monorimet@users.noreply.github.com> Date: Tue, 7 Jan 2025 11:02:29 -0600 Subject: [PATCH] (sdxl) Updates server configurations to run with caching allocator by default. (#768) This is a band-aid patch. We encountered a regression in under-load behavior (tracked by https://github.com/nod-ai/shark-ai/issues/753) This effectively uses a different allocator that, while less efficient at multi-device inference execution, is more stable. --- shortfin/python/shortfin_apps/sd/components/config_artifacts.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/shortfin/python/shortfin_apps/sd/components/config_artifacts.py b/shortfin/python/shortfin_apps/sd/components/config_artifacts.py index 432f08b4e..5c7875708 100644 --- a/shortfin/python/shortfin_apps/sd/components/config_artifacts.py +++ b/shortfin/python/shortfin_apps/sd/components/config_artifacts.py @@ -8,7 +8,7 @@ from iree.build.executor import FileNamespace import os -ARTIFACT_VERSION = "11182024" +ARTIFACT_VERSION = "01062025" SDXL_CONFIG_BUCKET = f"https://sharkpublic.blob.core.windows.net/sharkpublic/sdxl/{ARTIFACT_VERSION}/configs/"