Upgrade llama.cpp to b3265, support gemma2, remove beam parameter[ ggml-org/llama.cpp#7985 ]

zdl010 · kherud · commit eb77b904e0d2 · 2024-08-07T21:02:45.000+02:00
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -24,7 +24,7 @@ FetchContent_MakeAvailable(json)
 FetchContent_Declare(
 	llama.cpp
 	GIT_REPOSITORY https://github.com/ggerganov/llama.cpp.git
-	GIT_TAG        b3008
+	GIT_TAG        b3265
 )
 FetchContent_MakeAvailable(llama.cpp)
 
diff --git a/pom.xml b/pom.xml
@@ -4,7 +4,7 @@
 
 	<groupId>de.kherud</groupId>
 	<artifactId>llama</artifactId>
-	<version>3.2.1</version>
+	<version>3.2.2</version>
 	<packaging>jar</packaging>
 
 	<name>${project.groupId}:${project.artifactId}</name>
diff --git a/src/main/cpp/server.hpp b/src/main/cpp/server.hpp
@@ -2551,7 +2551,6 @@ static void server_params_parse(json jparams, server_params &sparams, gpt_params
     params.n_parallel = json_value(jparams, "n_parallel", default_params.n_parallel);
     params.n_sequences = json_value(jparams, "n_sequences", default_params.n_sequences);
     params.p_split = json_value(jparams, "p_split", default_params.p_split);
-    params.n_beams = json_value(jparams, "n_beams", default_params.n_beams);
     params.grp_attn_n = json_value(jparams, "grp_attn_n", default_params.grp_attn_n);
     params.grp_attn_w = json_value(jparams, "grp_attn_w", default_params.grp_attn_w);
     params.n_print = json_value(jparams, "n_print", default_params.n_print);
diff --git a/src/main/java/de/kherud/llama/ModelParameters.java b/src/main/java/de/kherud/llama/ModelParameters.java
@@ -32,7 +32,6 @@ public final class ModelParameters extends JsonParameters {
 	private static final String PARAM_SPLIT_MODE = "split_mode";
 	private static final String PARAM_MAIN_GPU = "main_gpu";
 	private static final String PARAM_TENSOR_SPLIT = "tensor_split";
-	private static final String PARAM_N_BEAMS = "n_beams";
 	private static final String PARAM_GRP_ATTN_N = "grp_attn_n";
 	private static final String PARAM_GRP_ATTN_W = "grp_attn_w";
 	private static final String PARAM_ROPE_FREQ_BASE = "rope_freq_base";
@@ -244,14 +243,6 @@ public ModelParameters setTensorSplit(float[] tensorSplit) {
 		return this;
 	}
 
-	/**
-	 * Set usage of beam search of given width if non-zero.
-	 */
-	public ModelParameters setNBeams(int nBeams) {
-		parameters.put(PARAM_N_BEAMS, String.valueOf(nBeams));
-		return this;
-	}
-
 	/**
 	 * Set the group-attention factor (default: 1)
 	 */

Original file line number	Diff line number	Diff line change
`@@ -24,7 +24,7 @@ FetchContent_MakeAvailable(json)`
`24`	`24`	`FetchContent_Declare(`
`25`	`25`	`llama.cpp`
`26`	`26`	`GIT_REPOSITORY https://github.com/ggerganov/llama.cpp.git`
`27`		`- GIT_TAG b3008`
	`27`	`+ GIT_TAG b3265`
`28`	`28`	`)`
`29`	`29`	`FetchContent_MakeAvailable(llama.cpp)`
`30`	`30`