From 14f2f3d935b05a4a3e998f365a317a5d8943b2cf Mon Sep 17 00:00:00 2001 From: Jake Luciani Date: Sat, 19 Oct 2024 13:13:33 -0400 Subject: [PATCH] Default to Q4 quantization --- .../com/github/tjake/jlama/cli/commands/QuantizeCommand.java | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/jlama-cli/src/main/java/com/github/tjake/jlama/cli/commands/QuantizeCommand.java b/jlama-cli/src/main/java/com/github/tjake/jlama/cli/commands/QuantizeCommand.java index 66d565e..a69e464 100644 --- a/jlama-cli/src/main/java/com/github/tjake/jlama/cli/commands/QuantizeCommand.java +++ b/jlama-cli/src/main/java/com/github/tjake/jlama/cli/commands/QuantizeCommand.java @@ -29,8 +29,8 @@ public class QuantizeCommand extends SimpleBaseCommand { @CommandLine.Parameters(index = "1", arity = "0..1", description = "The output location") protected Path output; - @CommandLine.Option(names = { "--quantization" }, paramLabel = "ARG", description = "Model quantization type", arity = "1") - protected DType modelQuantization; + @CommandLine.Option(names = { "--quantization" }, paramLabel = "ARG", description = "Model quantization type (default: ${DEFAULT-VALUE})", arity = "1", defaultValue = "Q4") + protected DType modelQuantization = DType.Q4; @CommandLine.Option(names = { "--skip-layer" }, paramLabel = "ARG", description = "Layer name prefix to not quantize") protected String[] skipLayerPrefixes;