fix: 模型修复token限制

kicc-platform/kicc-platform-biz/kicc-system-biz/src/main/java/com/cloud/kicc/system/service/impl/ImContentServiceImpl.java

@@ -18,7 +18,6 @@ import com.cloud.kicc.system.mapper.ImContentMapper;
 import com.cloud.kicc.system.service.FileService;
 import com.cloud.kicc.system.service.IImContentService;
 import com.cloud.kicc.system.util.AiUtil;
-import com.knuddels.jtokkit.api.ModelType;
 import com.pig4cloud.plugin.oss.OssProperties;
 import com.pig4cloud.plugin.oss.service.OssTemplate;
 import com.theokanning.openai.audio.CreateSpeechRequest;
@@ -29,7 +28,6 @@ import com.theokanning.openai.completion.chat.ChatCompletionResult;
 import com.theokanning.openai.completion.chat.ChatMessage;
 import com.theokanning.openai.completion.chat.ChatMessageRole;
 import com.theokanning.openai.service.OpenAiService;
-import com.theokanning.openai.utils.TikTokensUtil;
 import lombok.RequiredArgsConstructor;
 import lombok.SneakyThrows;
 import okhttp3.ResponseBody;
@@ -94,7 +92,7 @@ public class ImContentServiceImpl extends ServiceImpl<ImContentMapper, ImContent
                 i.eq(ImContent::getSendUserId, imContent.getReceiveUserId())
                  .eq(ImContent::getReceiveUserId, imContent.getSendUserId())
         ).last(String.format("LIMIT %s", openAiConfigProperties.getContextMessageMaxCount()))
-         .orderByAsc(ImContent::getSendTime));
+         .orderByDesc(ImContent::getSendTime));
 
         // 历史上下文消息
         List<ChatMessage> historyMessages = imContents.stream().map(item -> {
@@ -103,8 +101,6 @@ public class ImContentServiceImpl extends ServiceImpl<ImContentMapper, ImContent
             chatMessage.setContent(item.getContent());
             return chatMessage;
         }).collect(Collectors.toList());
-        // 截断支持的上下文消息列表到令牌限制
-        truncateToTokenLimit(openAiConfigProperties.getModel(), historyMessages);
 
         // Completion消息
         ChatMessage completionMessage = new ChatMessage(ChatMessageRole.USER.value(), imContent.getContent());
@@ -265,23 +261,4 @@ public class ImContentServiceImpl extends ServiceImpl<ImContentMapper, ImContent
         return String.format("请过 %s小时%s分钟%s秒 之后在试!",hours,minutes,seconds);
     }
 
-
-    /**
-     * 截断支持的上下文消息列表到令牌限制
-     * @param messages 上下文消息列表
-     * @param modelName 指定模型
-     */
-    private void truncateToTokenLimit(String modelName, List<ChatMessage> messages) {
-        ModelType modelType = ModelType.fromName(modelName)
-                .orElseThrow(() -> new CheckedException(String.format("找不到指定的:%s模型请检查配置!", modelName)));;
-        int sumTokens = TikTokensUtil.tokens(modelName, messages);
-        // 从前向后遍历消息，直到总 token 数在限制之内
-        while (!messages.isEmpty() && sumTokens > modelType.getMaxContextLength()) {
-            // 移除列表中的第一个消息
-            messages.remove(0);
-            // 重新计算总 token 数
-            sumTokens = TikTokensUtil.tokens(modelName, messages);
-        }
-    }
-
 }

kicc-platform/kicc-platform-biz/kicc-system-biz/src/main/java/com/cloud/kicc/system/util/AiUtil.java

@@ -1,14 +1,17 @@
 package com.cloud.kicc.system.util;
 
+import com.cloud.kicc.common.core.exception.CheckedException;
 import com.cloud.kicc.common.core.util.SpringContextHolderUtil;
 import com.cloud.kicc.common.security.util.SecurityUtils;
 import com.cloud.kicc.system.config.OpenAiConfigProperties;
+import com.knuddels.jtokkit.api.ModelType;
 import com.theokanning.openai.client.OpenAiApi;
 import com.theokanning.openai.completion.chat.ChatCompletionRequest;
 import com.theokanning.openai.completion.chat.ChatCompletionResult;
 import com.theokanning.openai.completion.chat.ChatMessage;
 import com.theokanning.openai.completion.chat.ChatMessageRole;
 import com.theokanning.openai.service.OpenAiService;
+import com.theokanning.openai.utils.TikTokensUtil;
 import lombok.experimental.UtilityClass;
 import okhttp3.OkHttpClient;
 import retrofit2.Retrofit;
@@ -56,6 +59,7 @@ public class AiUtil {
         ChatMessage systemMessage = new ChatMessage(ChatMessageRole.SYSTEM.value(), openAiConfigProperties.getSystemRule());
         messages.add(systemMessage);
         messages.addAll(historyMessages);
+        truncateToTokenLimit(openAiConfigProperties.getModel(), historyMessages);
         ChatCompletionRequest request = ChatCompletionRequest.builder()
                 .model(openAiConfigProperties.getModel())
                 .messages(messages)
@@ -69,4 +73,27 @@ public class AiUtil {
         return getOpenAiService().createChatCompletion(request);
     }
 
+
+    /**
+     * <a href="https://platform.openai.com/docs/models/gpt-4-and-gpt-4-turbo">...</a>
+     * 截断支持的上下文消息列表到令牌限制
+     * @param messages 上下文消息列表
+     * @param modelName 指定模型
+     */
+    public void truncateToTokenLimit(String modelName, List<ChatMessage> messages) {
+        ModelType modelType = ModelType.fromName(modelName)
+                .orElseThrow(() -> new CheckedException(String.format("找不到指定的:%s模型请检查配置!", modelName)));;
+        int sumTokens = TikTokensUtil.tokens(modelName, messages);
+        // 确保至少有一条消息（系统消息）保持不变
+        if (!messages.isEmpty()) {
+            // 从前向后遍历消息（从第二条开始），直到总 token 数在限制之内
+            while (messages.size() > 1 && sumTokens > modelType.getMaxContextLength()) {
+                // 移除列表中的第二条及以后的消息
+                messages.remove(1);
+                // 重新计算总 token 数
+                sumTokens = TikTokensUtil.tokens(modelName, messages);
+            }
+        }
+    }
+
 }