diff --git a/llamacpp/deployment-lfm25-26b.yaml b/llamacpp/deployment-lfm25-26b.yaml index 41fe4df..b6a2126 100644 --- a/llamacpp/deployment-lfm25-26b.yaml +++ b/llamacpp/deployment-lfm25-26b.yaml @@ -64,7 +64,7 @@ spec: - /models/LFM2.5-2.6B-Q4_0.gguf - --spec-default - --spec-type - - draft + - draft-simple - --ctx-size - "121000" - --temp