推进 ActionExtractor 语义缓存机制: 两种嵌入模型的连接方式测试完毕，在高性能主机上，可以通过ollama调用mxbai-embed-large这类模型，但放到4核8G香橙派3B就会出现推理时长过长，哪怕换成ONNX RUNTIME JAVA 也难以避免，但如果更换成 nomic-embed-text + ONNX RUNTIME JAVA ，仍能够拿到70左右ms的推理时长，远低于提取模型以及向量模型API的调用时长。预期可提供两种语义缓存所用的嵌入模型接入方式: 通过 http 调用 本地ollama接口; 指定 ONNX 格式的嵌入模型直接调用。

.idea/vcs.xml

@@ -2,6 +2,5 @@
 <project version="4">
   <component name="VcsDirectoryMappings">
     <mapping directory="$PROJECT_DIR$" vcs="Git" />
-    <mapping directory="$USER_HOME$/Projects/IdeaProjects/Projects/Partner" vcs="Git" />
   </component>
 </project>

.xcodemap/config/xcodemap-class-filter.yaml

@@ -1,33 +0,0 @@
-autoDetectedPackages:
-- factory
-- module
-- work.slhaf
-enableAutoDetect: true
-entryDisplayConfig:
-  excludedPathPatterns: []
-  skipJsCss: true
-funcDisplayConfig:
-  skipConstructors: false
-  skipFieldAccess: true
-  skipFieldChange: true
-  skipGetters: false
-  skipNonProjectPackages: false
-  skipPrivateMethods: false
-  skipSetters: false
-ignoreSameClassCall: null
-ignoreSamePackageCall: null
-includedPackagePrefixes: null
-includedParentClasses: null
-maxColSize: 32
-maxNumFirst: 12
-maxNumFirstImportant: 1024
-maxNumHash: 3
-maxNumHashImportant: 256
-maxObjectDepth: 4
-maxStrSize: 4096
-name: xcodemap-filter
-openMainWindow: true
-recordMode: manual
-sourceDisplayConfig:
-  color: blue
-startOnDebug: false

Partner-Main/pom.xml

@@ -34,6 +34,16 @@
             <artifactId>nd4j-api</artifactId>
             <version>1.0.0-M2.1</version>
         </dependency>
+        <dependency>
+            <groupId>com.microsoft.onnxruntime</groupId>
+            <artifactId>onnxruntime</artifactId>
+            <version>1.23.1</version>
+        </dependency>
+        <dependency>
+            <groupId>ai.djl.huggingface</groupId>
+            <artifactId>tokenizers</artifactId>
+            <version>0.34.0</version>
+        </dependency>
     </dependencies>
 
     <properties>

Partner-Main/src/main/java/work/slhaf/partner/module/modules/action/planner/ActionPlanner.java

@@ -8,8 +8,8 @@ import work.slhaf.partner.api.chat.pojo.Message;
 import work.slhaf.partner.common.thread.InteractionThreadPoolExecutor;
 import work.slhaf.partner.core.action.ActionCapability;
 import work.slhaf.partner.core.action.entity.*;
-import work.slhaf.partner.core.cache.CacheCapability;
 import work.slhaf.partner.core.cognation.CognationCapability;
+import work.slhaf.partner.core.memory.MemoryCapability;
 import work.slhaf.partner.core.perceive.PerceiveCapability;
 import work.slhaf.partner.module.common.module.PreRunningModule;
 import work.slhaf.partner.module.modules.action.planner.confirmer.ActionConfirmer;
@@ -42,7 +42,7 @@ public class ActionPlanner extends PreRunningModule {
     @InjectCapability
     private PerceiveCapability perceiveCapability;
     @InjectCapability
-    private CacheCapability cacheCapability;
+    private MemoryCapability memoryCapability;
 
     @InjectModule
     private ActionEvaluator actionEvaluator;
@@ -197,7 +197,7 @@ public class ActionPlanner extends PreRunningModule {
             input.setTendencies(extractorResult.getTendencies());
             input.setUser(perceiveCapability.getUser(userId));
             input.setRecentMessages(cognationCapability.getChatMessages());
-            input.setActivatedSlices(cacheCapability.getActivatedSlices(userId));
+            input.setActivatedSlices(memoryCapability.getActivatedSlices(userId));
             return input;
         }
 

Partner-Main/src/main/java/work/slhaf/partner/module/modules/memory/selector/MemorySelector.java

@@ -133,8 +133,10 @@ public class MemorySelector extends PreRunningModule {
         return "[记忆模块]";
     }
 
-    protected HashMap<String, String> getPromptDataMap(String userId) {
+    @Override
+    protected HashMap<String, String> getPromptDataMap(PartnerRunningFlowContext context) {
         HashMap<String, String> map = new HashMap<>();
+        String userId = context.getUserId();
         String dialogMapStr = memoryCapability.getDialogMapStr();
         if (!dialogMapStr.isEmpty()) {
             map.put("[记忆缓存] <你最近两日和所有聊天者的对话记忆印象>", dialogMapStr);

Partner-Main/src/test/java/OnnxTest.java

@@ -0,0 +1,103 @@
+import ai.djl.huggingface.tokenizers.Encoding;
+import ai.djl.huggingface.tokenizers.HuggingFaceTokenizer;
+import ai.onnxruntime.OnnxTensor;
+import ai.onnxruntime.OrtEnvironment;
+import ai.onnxruntime.OrtException;
+import ai.onnxruntime.OrtSession;
+import lombok.extern.slf4j.Slf4j;
+import org.junit.jupiter.api.BeforeAll;
+import org.junit.jupiter.api.Test;
+
+import java.io.IOException;
+import java.nio.file.Path;
+import java.util.Arrays;
+import java.util.HashMap;
+import java.util.Map;
+
+@Slf4j
+public class OnnxTest {
+    static String tokenizer_json;
+    static String base;
+    static String model;
+
+    @BeforeAll
+    static void init() {
+        base = "/home/slhaf/IdeaProjects/Projects/Partner/data/vector/";
+        tokenizer_json = base + "tokenizer.json";
+        model = base + "model_quantized.onnx";
+    }
+
+    @Test
+    void tokenizerTest() throws IOException {
+        long l1 = System.currentTimeMillis();
+        HuggingFaceTokenizer tokenizer = HuggingFaceTokenizer.newInstance(Path.of(tokenizer_json));
+        long l2 = System.currentTimeMillis();
+        Encoding encode = tokenizer.encode("test: Hello World");
+        long l3 = System.currentTimeMillis();
+        long[] ids = encode.getIds();
+        long[] attentionMask = encode.getAttentionMask();
+        log.info(Arrays.toString(ids));
+        log.info("-----------------------------");
+        log.info(Arrays.toString(attentionMask));
+        log.info("-----------------------------");
+        log.info("加载耗时: {}ms", l2 - l1);
+        log.info("计算耗时: {}ms", l3 - l2);
+        tokenizer.close();
+        /* 输出:
+         * [101, 3231, 1024, 7592, 2088, 102]
+         * -----------------------------
+         * [1, 1, 1, 1, 1, 1]
+         * -----------------------------
+         * 加载耗时: 4206ms
+         * 计算耗时: 1ms
+         */
+    }
+
+    @Test
+    void onnxTest() throws IOException, OrtException {
+        long l1 = System.currentTimeMillis();
+        HuggingFaceTokenizer tokenizer = HuggingFaceTokenizer.newInstance(Path.of(tokenizer_json));
+        long l2 = System.currentTimeMillis();//tokenizer加载耗时
+        Encoding encode = tokenizer.encode("test: Hello World");
+        long l3 = System.currentTimeMillis();//计算耗时
+
+        long[] ids = encode.getIds();
+        long[] attentionMask = encode.getAttentionMask();
+
+        long l4 = System.currentTimeMillis();
+        OrtEnvironment env = OrtEnvironment.getEnvironment();
+        OrtSession.SessionOptions ops = new OrtSession.SessionOptions();
+        OrtSession session = env.createSession(model, ops);
+        long l5 = System.currentTimeMillis();//模型加载耗时
+
+        long[][] inputIdsBatch = {ids};
+        long[][] attentionMaskBatch = {attentionMask};
+        long[][] tokenTypeIdsBatch = {new long[ids.length]};  // 初始化全 0
+        for (int i = 0; i < ids.length; i++) tokenTypeIdsBatch[0][i] = 0;
+
+        OnnxTensor inputTensor = OnnxTensor.createTensor(env, inputIdsBatch);
+        OnnxTensor maskTensor = OnnxTensor.createTensor(env, attentionMaskBatch);
+        OnnxTensor tokenTypeTensor = OnnxTensor.createTensor(env, tokenTypeIdsBatch);
+
+        Map<String, OnnxTensor> inputs = new HashMap<>();
+        inputs.put("input_ids", inputTensor);
+        inputs.put("attention_mask", maskTensor);
+        inputs.put("token_type_ids", tokenTypeTensor);
+
+        long l6 = System.currentTimeMillis();
+        OrtSession.Result result = session.run(inputs);
+        long l7 = System.currentTimeMillis();//模型计算耗时
+        OnnxTensor embeddingTensor = (OnnxTensor) result.get(0);
+        float[] embeddings = embeddingTensor.getFloatBuffer().array();
+
+
+        log.info(Arrays.toString(embeddings));
+        log.info("------------------------");
+        log.info("tokenizer加载耗时: {}ms", l2 - l1);
+        log.info("tokenizer计算耗时: {}ms", l3 - l2);
+        log.info("模型加载耗时: {}ms", l5 - l4);
+        log.info("模型数据准备耗时: {}ms", l6 - l5);
+        log.info("模型计算耗时: {}ms", l7 - l6);
+        tokenizer.close();
+    }
+}