Compare commits

..

2 Commits

2 changed files with 4 additions and 0 deletions

View File

@@ -232,6 +232,7 @@
jq
nasctui
trilium-desktop
haruna
];
fonts.packages = with pkgs; [

View File

@@ -90,6 +90,9 @@
# As long as this is here the models are declarative. llama-server will grab them if not downloaded already.
environment.etc."llama-swap/config.yaml".text = ''
models:
"Qwen3.6-35B-A3B-GGUF":
cmd: "llama-server --port ''${PORT} -hf unsloth/Qwen3.6-35B-A3B-GGUF:UD-Q4_K_S --ctx-size 128000 --temp 0.6 --top-p 0.95 --top-k 20 --min-p 0.00 -np 1 --fit-target 256 --ubatch-size 1024 -fa on --slots --slot-save-path /home/laythe/llamapcache --jinja -kvu --no-mmproj --swa-checkpoints 32 --no-kv-offload --chat-template-kwargs '{\"preserve_thinking\": true}'"
ttl: 2400
"Qwen3.5-35B-A3B-GGUF":
cmd: llama-server --port ''${PORT} -hf mudler/Qwen3.5-35B-A3B-APEX-GGUF:Mini --ctx-size 128000 --temp 0.6 --top-p 0.95 --top-k 20 --min-p 0.00 -np 1 --fit-target 256 --ubatch-size 1024 -fa on --slots --slot-save-path /home/laythe/llamapcache --jinja -kvu --no-mmproj --swa-checkpoints 32 --no-kv-offload
ttl: 2400