llm: use vulkan
This commit is contained in:
parent
870093686a
commit
fb4043712e
@ -4,6 +4,7 @@
|
|||||||
config,
|
config,
|
||||||
inputs,
|
inputs,
|
||||||
optimizePackage,
|
optimizePackage,
|
||||||
|
lib,
|
||||||
...
|
...
|
||||||
}:
|
}:
|
||||||
{
|
{
|
||||||
@ -17,20 +18,17 @@
|
|||||||
);
|
);
|
||||||
port = service_configs.ports.llama_cpp;
|
port = service_configs.ports.llama_cpp;
|
||||||
host = "0.0.0.0";
|
host = "0.0.0.0";
|
||||||
package = (
|
package = (optimizePackage inputs.llamacpp.packages.${pkgs.system}.vulkan);
|
||||||
optimizePackage (
|
|
||||||
inputs.llamacpp.packages.${pkgs.system}.default.overrideAttrs (old: {
|
|
||||||
cmakeFlags = old.cmakeFlags ++ [
|
|
||||||
"-DGGML_AVX2=ON"
|
|
||||||
];
|
|
||||||
})
|
|
||||||
)
|
|
||||||
);
|
|
||||||
extraFlags = [
|
extraFlags = [
|
||||||
"--flash-attn"
|
"--flash-attn"
|
||||||
|
"-ngl"
|
||||||
|
"9999"
|
||||||
];
|
];
|
||||||
};
|
};
|
||||||
|
|
||||||
|
# have to do this in order to get vulkan to work
|
||||||
|
systemd.services.llama-cpp.serviceConfig.DynamicUser = lib.mkForce false;
|
||||||
|
|
||||||
services.caddy.virtualHosts."llm.${service_configs.https.domain}".extraConfig = ''
|
services.caddy.virtualHosts."llm.${service_configs.https.domain}".extraConfig = ''
|
||||||
${builtins.readFile ../secrets/caddy_auth}
|
${builtins.readFile ../secrets/caddy_auth}
|
||||||
reverse_proxy :${builtins.toString config.services.llama-cpp.port}
|
reverse_proxy :${builtins.toString config.services.llama-cpp.port}
|
||||||
|
|||||||
Loading…
x
Reference in New Issue
Block a user