(improvement)(common) Fix the issue with querying large models using the GLM interface. (#1226)

2025-12-14 13:47:09 +00:00 · 2024-06-25 21:42:02 +08:00
parent 8cf5380cb8
commit e9dfb30ccf
2 changed files with 378 additions and 0 deletions
--- a/common/src/main/java/dev/langchain4j/model/zhipu/ZhipuAiChatModel.java
+++ b/common/src/main/java/dev/langchain4j/model/zhipu/ZhipuAiChatModel.java
@@ -0,0 +1,114 @@
+package dev.langchain4j.model.zhipu;
+
+import dev.ai4j.openai4j.chat.ChatCompletionModel;
+import dev.langchain4j.agent.tool.ToolSpecification;
+import dev.langchain4j.data.message.AiMessage;
+import dev.langchain4j.data.message.ChatMessage;
+import dev.langchain4j.model.chat.ChatLanguageModel;
+import dev.langchain4j.model.output.Response;
+import dev.langchain4j.model.zhipu.chat.ChatCompletionRequest;
+import dev.langchain4j.model.zhipu.chat.ChatCompletionResponse;
+import dev.langchain4j.model.zhipu.spi.ZhipuAiChatModelBuilderFactory;
+import lombok.Builder;
+
+import java.util.List;
+
+import static dev.langchain4j.internal.RetryUtils.withRetry;
+import static dev.langchain4j.internal.Utils.getOrDefault;
+import static dev.langchain4j.internal.Utils.isNullOrEmpty;
+import static dev.langchain4j.internal.ValidationUtils.ensureNotEmpty;
+import static dev.langchain4j.model.zhipu.DefaultZhipuAiHelper.aiMessageFrom;
+import static dev.langchain4j.model.zhipu.DefaultZhipuAiHelper.finishReasonFrom;
+import static dev.langchain4j.model.zhipu.DefaultZhipuAiHelper.toTools;
+import static dev.langchain4j.model.zhipu.DefaultZhipuAiHelper.toZhipuAiMessages;
+import static dev.langchain4j.model.zhipu.DefaultZhipuAiHelper.tokenUsageFrom;
+import static dev.langchain4j.model.zhipu.chat.ToolChoiceMode.AUTO;
+import static dev.langchain4j.spi.ServiceHelper.loadFactories;
+import static java.util.Collections.singletonList;
+
+/**
+ * Represents an ZhipuAi language model with a chat completion interface, such as glm-3-turbo and glm-4.
+ * You can find description of parameters <a href="https://open.bigmodel.cn/dev/api">here</a>.
+ */
+public class ZhipuAiChatModel implements ChatLanguageModel {
+
+    private final String baseUrl;
+    private final Double temperature;
+    private final Double topP;
+    private final String model;
+    private final Integer maxRetries;
+    private final Integer maxToken;
+    private final ZhipuAiClient client;
+
+    @Builder
+    public ZhipuAiChatModel(
+            String baseUrl,
+            String apiKey,
+            Double temperature,
+            Double topP,
+            String model,
+            Integer maxRetries,
+            Integer maxToken,
+            Boolean logRequests,
+            Boolean logResponses
+    ) {
+        this.baseUrl = getOrDefault(baseUrl, "https://open.bigmodel.cn/");
+        this.temperature = getOrDefault(temperature, 0.7);
+        this.topP = topP;
+        this.model = getOrDefault(model, ChatCompletionModel.GPT_4.toString());
+        this.maxRetries = getOrDefault(maxRetries, 3);
+        this.maxToken = getOrDefault(maxToken, 512);
+        this.client = ZhipuAiClient.builder()
+                .baseUrl(this.baseUrl)
+                .apiKey(apiKey)
+                .logRequests(getOrDefault(logRequests, false))
+                .logResponses(getOrDefault(logResponses, false))
+                .build();
+    }
+
+    public static ZhipuAiChatModelBuilder builder() {
+        for (ZhipuAiChatModelBuilderFactory factories : loadFactories(ZhipuAiChatModelBuilderFactory.class)) {
+            return factories.get();
+        }
+        return new ZhipuAiChatModelBuilder();
+    }
+
+    @Override
+    public Response<AiMessage> generate(List<ChatMessage> messages) {
+        return generate(messages, (ToolSpecification) null);
+    }
+
+    @Override
+    public Response<AiMessage> generate(List<ChatMessage> messages, List<ToolSpecification> toolSpecifications) {
+        ensureNotEmpty(messages, "messages");
+
+        ChatCompletionRequest.Builder requestBuilder = ChatCompletionRequest.builder()
+                .model(this.model)
+                .maxTokens(maxToken)
+                .stream(false)
+                .topP(topP)
+                .toolChoice(AUTO)
+                .messages(toZhipuAiMessages(messages));
+
+        if (!isNullOrEmpty(toolSpecifications)) {
+            requestBuilder.tools(toTools(toolSpecifications));
+        }
+
+        ChatCompletionResponse response = withRetry(() -> client.chatCompletion(requestBuilder.build()), maxRetries);
+        return Response.from(
+                aiMessageFrom(response),
+                tokenUsageFrom(response.getUsage()),
+                finishReasonFrom(response.getChoices().get(0).getFinishReason())
+        );
+    }
+
+    @Override
+    public Response<AiMessage> generate(List<ChatMessage> messages, ToolSpecification toolSpecification) {
+        return generate(messages, toolSpecification != null ? singletonList(toolSpecification) : null);
+    }
+
+    public static class ZhipuAiChatModelBuilder {
+        public ZhipuAiChatModelBuilder() {
+        }
+    }
+}