ConfigurableLanguageModel implements Generator

vespa-engine · Nov 29, 2024 · f4869ec · f4869ec
1 parent 0f0d6a9
commit f4869ec
Show file tree

Hide file tree

Showing 5 changed files with 22 additions and 15 deletions.
diff --git a/model-integration/abi-spec.json b/model-integration/abi-spec.json
@@ -2,7 +2,8 @@
   "ai.vespa.llm.clients.ConfigurableLanguageModel" : {
     "superClass" : "java.lang.Object",
     "interfaces" : [
-      "ai.vespa.llm.LanguageModel"
+      "ai.vespa.llm.LanguageModel",
+      "com.yahoo.language.process.Generator"
     ],
     "attributes" : [
       "public",
@@ -14,7 +15,8 @@
       "protected java.lang.String getApiKey(ai.vespa.llm.InferenceParameters)",
       "protected void setApiKey(ai.vespa.llm.InferenceParameters)",
       "protected java.lang.String getEndpoint()",
-      "protected void setEndpoint(ai.vespa.llm.InferenceParameters)"
+      "protected void setEndpoint(ai.vespa.llm.InferenceParameters)",
+      "public java.lang.String generate(ai.vespa.llm.completion.Prompt, com.yahoo.language.process.Generator$Context)"
     ],
     "fields" : [ ]
   },
@@ -174,17 +176,14 @@
   },
   "ai.vespa.llm.clients.OpenAI" : {
     "superClass" : "ai.vespa.llm.clients.ConfigurableLanguageModel",
-    "interfaces" : [
-      "com.yahoo.language.process.Generator"
-    ],
+    "interfaces" : [ ],
     "attributes" : [
       "public"
     ],
     "methods" : [
       "public void <init>(ai.vespa.llm.clients.LlmClientConfig, ai.vespa.secret.Secrets)",
       "public java.util.List complete(ai.vespa.llm.completion.Prompt, ai.vespa.llm.InferenceParameters)",
-      "public java.util.concurrent.CompletableFuture completeAsync(ai.vespa.llm.completion.Prompt, ai.vespa.llm.InferenceParameters, java.util.function.Consumer)",
-      "public java.lang.String generate(ai.vespa.llm.completion.Prompt, com.yahoo.language.process.Generator$Context)"
+      "public java.util.concurrent.CompletableFuture completeAsync(ai.vespa.llm.completion.Prompt, ai.vespa.llm.InferenceParameters, java.util.function.Consumer)"
     ],
     "fields" : [ ]
   },

diff --git a/model-integration/src/main/java/ai/vespa/llm/clients/ConfigurableLanguageModel.java b/model-integration/src/main/java/ai/vespa/llm/clients/ConfigurableLanguageModel.java
@@ -8,6 +8,7 @@
 import ai.vespa.secret.Secrets;
 import com.yahoo.api.annotations.Beta;
 import com.yahoo.component.annotation.Inject;
+import com.yahoo.language.process.Generator;
 
 import java.util.HashMap;
 import java.util.logging.Logger;
@@ -19,7 +20,7 @@
  * @author lesters
  */
 @Beta
-public abstract class ConfigurableLanguageModel implements LanguageModel {
+public abstract class ConfigurableLanguageModel implements LanguageModel, Generator {
 
     private static final Logger log = Logger.getLogger(ConfigurableLanguageModel.class.getName());
 
@@ -77,4 +78,15 @@ protected void setEndpoint(InferenceParameters params) {
         }
     }
 
+    @Override
+    public String generate(Prompt prompt, Context context) {
+        var params = new HashMap<String, String>();
+        var options = new InferenceParameters(params::get);
+        setApiKey(options);
+
+        var completions = complete(prompt, options);
+        var firstCompletion = completions.get(0);
+        return firstCompletion.text();
+    }
+
 }
diff --git a/model-integration/src/main/java/ai/vespa/llm/clients/LocalLLM.java b/model-integration/src/main/java/ai/vespa/llm/clients/LocalLLM.java
@@ -9,7 +9,6 @@
 import ai.vespa.llm.completion.Prompt;
 import com.yahoo.component.AbstractComponent;
 import com.yahoo.component.annotation.Inject;
-import com.yahoo.document.DataType;
 import com.yahoo.language.process.Generator;
 import de.kherud.llama.LlamaModel;
 import de.kherud.llama.ModelParameters;

diff --git a/model-integration/src/main/java/ai/vespa/llm/clients/OpenAI.java b/model-integration/src/main/java/ai/vespa/llm/clients/OpenAI.java
@@ -21,7 +21,7 @@
  * @author lesters
  */
 @Beta
-public class OpenAI extends ConfigurableLanguageModel implements Generator {
+public class OpenAI extends ConfigurableLanguageModel {
 
     private final OpenAiClient client;
 
@@ -47,10 +47,5 @@ public CompletableFuture<Completion.FinishReason> completeAsync(Prompt prompt,
         return client.completeAsync(prompt, parameters, consumer);
     }
 
-    @Override
-    public String generate(Prompt prompt, Context context) {
-        return GeneratorUtils.generate(prompt, this);
-    }
-
 }
 
diff --git a/quickbuild.sh b/quickbuild.sh
@@ -0,0 +1,2 @@
+mvn clean install --threads 1C -Dmaven.javadoc.skip=true -Dmaven.source.skip=true -DskipTests "$@"
+
Original file line number	Diff line number	Diff line change
		@@ -0,0 +1,2 @@
		mvn clean install --threads 1C -Dmaven.javadoc.skip=true -Dmaven.source.skip=true -DskipTests "$@"