diff options
| author | Franck Cuny <franck@fcuny.net> | 2026-01-03 11:26:18 -0800 |
|---|---|---|
| committer | Franck Cuny <franck@fcuny.net> | 2026-01-03 11:26:18 -0800 |
| commit | 367b88492c7e3e49f61d2a5336c49a3bb308e9ff (patch) | |
| tree | 0ac55c5624ed640bf9cb99676167aece40a030d4 /profiles | |
| parent | fix typo (diff) | |
| download | infra-367b88492c7e3e49f61d2a5336c49a3bb308e9ff.tar.gz | |
delete ollama
Diffstat (limited to '')
| -rw-r--r-- | profiles/llm.nix | 18 |
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 - ]; - }; -} |
