marvin: update llama.cpp config
2 files changed, 58 insertions(+), 20 deletions(-)
changed files
M user/hosts/marvin.nix → user/hosts/marvin.nix
@@ -89,25 +89,28 @@ }; }; llama-server = { enable = true; - config = { - ProgramArguments = [ - "${pkgs.llama-cpp}/bin/llama-server" - "--port" - "9090" - "--models-preset" - "${config.xdg.configHome}/llama/presets.ini" - ]; - RunAtLoad = true; - KeepAlive = true; - WorkingDirectory = "/Users/alan"; - StandardOutPath = "/Users/alan/Library/Logs/llama-server.log"; - StandardErrorPath = "/Users/alan/Library/Logs/llama-server.log"; - }; + config = + let + presets = "${nixfiles}/user/settings/llama/presets.ini"; + in + { + ProgramArguments = [ + "${pkgs.llama-cpp}/bin/llama-server" + "--port" + "9090" + "--models-preset" + "${presets}" + ]; + RunAtLoad = true; + KeepAlive = true; + WorkingDirectory = "/Users/alan"; + StandardOutPath = "/Users/alan/Library/Logs/llama-server.log"; + StandardErrorPath = "/Users/alan/Library/Logs/llama-server.log"; + WatchPaths = [ + "${presets}" + ]; + }; }; - }; - - xdg.configFile."llama/presets.ini" = { - source = direct_symlink "/user/settings/llama/presets.ini"; }; xdg.configFile."raycast/ai/providers.yaml" = {
M user/settings/llama/presets.ini → user/settings/llama/presets.ini
@@ -1,10 +1,31 @@ version = 1 -[ggml-org/gpt-oss-20b-GGUF] +[unsloth/gpt-oss-20b-GGUF:Q8_0] jinja = on +temp = 1.0 +top-k = 0.0 +top-p = 1.0 +min-p = 0.0 +load-on-startup = on +chat-template-kwargs = {"reasoning_effort": "high"} -[ggml-org/gpt-oss-120b-GGUF] +[unsloth/gpt-oss-120b-GGUF:UD-Q4_K_XL] jinja = on +load-on-startup = on +temp = 1.0 +top-k = 0.0 +top-p = 1.0 +min-p = 0.0 +chat-template-kwargs = {"reasoning_effort": "high"} + +[unsloth/gemma-3n-E4B-it-GGUF:Q8_0] +jinja = on +load-on-startup = on +ctx-size = 32768 +temp = 1.0 +top-k = 64 +top-p = 0.95 +min-p = 0.0 [henry1477/Qwen3-Embedding-4B-gguf] embedding = on@@ -12,7 +33,21 @@ pooling = cls ubatch-size = 8192 [ggml-org/Ministral-3-14B-Reasoning-2512-GGUF] +temp = 0.7 +top-p = 0.95 [ggml-org/Ministral-3-14B-Instruct-2512-GGUF] +temp = 0.1 [ggml-org/Nemotron-Nano-3-30B-A3B-GGUF] +jinja = on +ctx-size = 1048576 +fit = on +special = on +load-on-startup = on + +[unsloth/MiniMax-M2.1-GGUF:Q3_K_XL] +jinja = on + +[unsloth/SmolLM2-1.7B-Instruct-GGUF:Q8_0] +jinja = on