marvin: use llama-server's routing instead of llama-swap
1 file changed, 18 insertions(+), 0 deletions(-)
changed files
A user/settings/llama/presets.ini
@@ -0,0 +1,18 @@ +version = 1 + +[ggml-org/gpt-oss-20b-GGUF] +jinja = on + +[ggml-org/gpt-oss-120b-GGUF] +jinja = on + +[henry1477/Qwen3-Embedding-4B-gguf] +embedding = on +pooling = cls +ubatch-size = 8192 + +[ggml-org/Ministral-3-14B-Reasoning-2512-GGUF] + +[ggml-org/Ministral-3-14B-Instruct-2512-GGUF] + +[ggml-org/Nemotron-Nano-3-30B-A3B-GGUF]