diff --git a/dev-phi-3-mini.yaml b/dev-phi-3-mini.yaml index 32be9eb974f2d8094f8bbc59d2a8ae335afbc3f7..d4b99351a00718f1b183e9e7d81b8c7ad8f8af64 100644 --- a/dev-phi-3-mini.yaml +++ b/dev-phi-3-mini.yaml @@ -12,7 +12,7 @@ config_file: | threads: 64 gpu_layers: 35 parameters: - model: "downloads/llama_backend/Phi-3.1-mini-128k-instruct-Q6_K_L.gguf" + model: "downloads/llama_backend/Phi-3.5-mini-instruct-Q6_K.gguf" template: chat_message: | <|{{ .RoleName }}|> @@ -26,8 +26,9 @@ config_file: | - <|end|> files: -- filename: "downloads/llama_backend/Phi-3.1-mini-128k-instruct-Q6_K_L.gguf" +- filename: "downloads/llama_backend/Phi-3.5-mini-instruct-Q6_K.gguf" sha256: "" - uri: "https://huggingface.co/bartowski/Phi-3.1-mini-128k-instruct-GGUF/resolve/main/Phi-3.1-mini-128k-instruct-Q6_K_L.gguf" + uri: "https://huggingface.co/bartowski/Phi-3.5-mini-instruct-GGUF/resolve/main/Phi-3.5-mini-instruct-Q6_K.gguf" + # uri: "https://huggingface.co/bartowski/Phi-3.1-mini-128k-instruct-GGUF/resolve/main/Phi-3.1-mini-128k-instruct-Q6_K_L.gguf"