From 414fb71a8454e56142336d572443901f9c60044c Mon Sep 17 00:00:00 2001
From: "Chris (wolcen) Thompson" <chris@agaric.coop>
Date: Tue, 3 Feb 2026 16:55:59 -0500
Subject: [PATCH] Adjust ollama settings

---
 hosts/default/configuration.nix | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/hosts/default/configuration.nix b/hosts/default/configuration.nix
index 145e5e9..fad2859 100644
--- a/hosts/default/configuration.nix
+++ b/hosts/default/configuration.nix
@@ -433,6 +433,12 @@ in
     environmentVariables = {
       HCC_AMDGPU_TARGET = "gfx1031"; # used to be necessary, but doesn't seem to anymore
       OLLAMA_LOAD_TIMEOUT = "2m"; # Reduce load timeout (from 5 min)...if it's that big, forget it.
+      OLLAMA_CONTEXT_LENGTH = "65536";
+      OLLAMA_KEEP_ALIVE = "10m";
+      OLLAMA_MAX_LOADED_MODELS = "1";
+      OLLAMA_KV_CACHE_TYPE = "Q4_K_0";
+      OLLAMA_DEBUG = "1";
+      OLLAMA_FLASH_ATTENTION = "true";
       # OK, so origins is more about setting up CORS than firewalling things. (and requires e.g. http[s]://IP|Host/* etc)
       #OLLAMA_ORIGINS = "10.40.4.2"; # Only accept connections from locally.