marvin/dev: set up llama.cpp (via llama-swap) Faster inference than ollama, at least with gpt-oss-20b (around 20%)
1 file changed, 12 insertions(+), 0 deletions(-)
changed files
M system/settings/dev.nix → system/settings/dev.nix
@@ -118,6 +118,18 @@ } } ''; }; + "swap.localhost" = { + logFormat = "output discard"; + extraConfig = '' + import local_tls + reverse_proxy http://localhost:9292 { + transport http { + dial_timeout 1s + compression off + } + } + ''; + }; }; }; };