aboutsummaryrefslogtreecommitdiff
path: root/profiles/llm.nix
diff options
context:
space:
mode:
authorFranck Cuny <franck@fcuny.net>2026-01-03 11:26:18 -0800
committerFranck Cuny <franck@fcuny.net>2026-01-03 11:26:18 -0800
commit367b88492c7e3e49f61d2a5336c49a3bb308e9ff (patch)
tree0ac55c5624ed640bf9cb99676167aece40a030d4 /profiles/llm.nix
parentfix typo (diff)
downloadinfra-367b88492c7e3e49f61d2a5336c49a3bb308e9ff.tar.gz
delete ollama
Diffstat (limited to 'profiles/llm.nix')
-rw-r--r--profiles/llm.nix18
1 files changed, 0 insertions, 18 deletions
diff --git a/profiles/llm.nix b/profiles/llm.nix
deleted file mode 100644
index d5d02f4..0000000
--- a/profiles/llm.nix
+++ /dev/null
@@ -1,18 +0,0 @@
-{ ... }:
-{
- services.ollama = {
- enable = true;
- acceleration = "rocm";
- host = "0.0.0.0";
- openFirewall = true;
- environmentVariables = {
- # Flash Attention: Improves memory efficiency and speeds up long context
- OLLAMA_FLASH_ATTENTION = "1";
- # KV cache quantization: Reduces VRAM footprint (q8_0 = 8-bit quantization)
- OLLAMA_KV_CACHE_TYPE = "q8_0";
- };
- loadModels = [
- "qwen3:8b" # https://ollama.com/library/qwen3
- ];
- };
-}