From da88a9b2d66498f9832518a6297f48d244f51005 Mon Sep 17 00:00:00 2001
From: Harald Hoyer <harald@hoyer.xyz>
Date: Tue, 5 May 2026 10:41:05 +0200
Subject: [PATCH] fix(halo): drop speculative HSA_OVERRIDE_GFX_VERSION from
 llama-server

Was set defensively without knowing the actual GPU arch; if ROCm
supports the card natively, the override is at best a no-op and at
worst masks the real arch. Add it back with the right value if the
service actually fails to detect the GPU.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---
 systems/x86_64-linux/halo/llama-server.nix | 1 -
 1 file changed, 1 deletion(-)

diff --git a/systems/x86_64-linux/halo/llama-server.nix b/systems/x86_64-linux/halo/llama-server.nix
index e55e08c..9e655f1 100644
--- a/systems/x86_64-linux/halo/llama-server.nix
+++ b/systems/x86_64-linux/halo/llama-server.nix
@@ -13,7 +13,6 @@
     environment = {
       HOME = "%S/llama-server";
       HF_HOME = "%S/llama-server";
-      HSA_OVERRIDE_GFX_VERSION = lib.mkDefault "11.0.0";
     };
 
     serviceConfig = {