#
Junjie
昨天 ba0dae92a0eb14b378fd44c91ed2ccb481c164e3
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
package com.zy.ai.service;
 
import com.zy.ai.entity.ChatCompletionRequest;
import com.zy.ai.entity.ChatCompletionResponse;
import lombok.RequiredArgsConstructor;
import lombok.extern.slf4j.Slf4j;
import org.springframework.beans.factory.annotation.Value;
import org.springframework.http.HttpHeaders;
import org.springframework.http.MediaType;
import org.springframework.stereotype.Service;
import org.springframework.web.reactive.function.client.WebClient;
import reactor.core.publisher.Mono;
import reactor.core.publisher.Flux;
 
import java.util.List;
import java.util.function.Consumer;
import java.util.concurrent.LinkedBlockingQueue;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.atomic.AtomicBoolean;
import com.alibaba.fastjson.JSON;
import com.alibaba.fastjson.JSONArray;
import com.alibaba.fastjson.JSONObject;
 
@Slf4j
@Service
@RequiredArgsConstructor
public class LlmChatService {
 
    private final WebClient llmWebClient;
 
    @Value("${llm.api-key}")
    private String apiKey;
 
    @Value("${llm.model}")
    private String model;
 
    /**
     * 通用对话方法:传入 messages,返回大模型文本回复
     */
    public String chat(List<ChatCompletionRequest.Message> messages,
                       Double temperature,
                       Integer maxTokens) {
 
        ChatCompletionRequest req = new ChatCompletionRequest();
        req.setModel(model);
        req.setMessages(messages);
        req.setTemperature(temperature != null ? temperature : 0.3);
        req.setMax_tokens(maxTokens != null ? maxTokens : 1024);
 
        ChatCompletionResponse response = llmWebClient.post()
                .uri("/chat/completions")
                .header(HttpHeaders.AUTHORIZATION, "Bearer " + apiKey)
                .contentType(MediaType.APPLICATION_JSON)
                .bodyValue(req)
                .retrieve()
                .bodyToMono(ChatCompletionResponse.class)
                .doOnError(ex -> log.error("调用 LLM 失败", ex))
                .onErrorResume(ex -> Mono.empty())
                .block();
 
        if (response == null ||
                response.getChoices() == null ||
                response.getChoices().isEmpty() ||
                response.getChoices().get(0).getMessage() == null) {
 
            return "AI 诊断失败:未获取到有效回复。";
        }
 
        return response.getChoices().get(0).getMessage().getContent();
    }
 
    public void chatStream(List<ChatCompletionRequest.Message> messages,
                           Double temperature,
                           Integer maxTokens,
                           Consumer<String> onChunk,
                           Runnable onComplete,
                           Consumer<Throwable> onError) {
 
        ChatCompletionRequest req = new ChatCompletionRequest();
        req.setModel(model);
        req.setMessages(messages);
        req.setTemperature(temperature != null ? temperature : 0.3);
        req.setMax_tokens(maxTokens != null ? maxTokens : 1024);
        req.setStream(true);
 
 
        Flux<String> flux = llmWebClient.post()
                .uri("/chat/completions")
                .header(HttpHeaders.AUTHORIZATION, "Bearer " + apiKey)
                .contentType(MediaType.APPLICATION_JSON)
                .accept(MediaType.TEXT_EVENT_STREAM)
                .bodyValue(req)
                .retrieve()
                .bodyToFlux(String.class)
                .doOnError(ex -> log.error("调用 LLM 流式失败", ex));
 
        AtomicBoolean doneSeen = new AtomicBoolean(false);
        AtomicBoolean errorSeen = new AtomicBoolean(false);
        LinkedBlockingQueue<String> queue = new LinkedBlockingQueue<>();
 
        Thread drain = new Thread(() -> {
            try {
                while (true) {
                    String s = queue.poll(2, TimeUnit.SECONDS);
                    if (s != null) {
                        try { onChunk.accept(s); } catch (Exception ignore) {}
                    }
                    if (doneSeen.get() && queue.isEmpty()) {
                        if (!errorSeen.get()) {
                            try { if (onComplete != null) onComplete.run(); } catch (Exception ignore) {}
                        }
                        break;
                    }
                }
            } catch (InterruptedException ignore) {
                ignore.printStackTrace();
            }
        });
        drain.setDaemon(true);
        drain.start();
 
        flux.subscribe(payload -> {
            if (payload == null || payload.isEmpty()) return;
            String[] events = payload.split("\\r?\\n\\r?\\n");
            for (String part : events) {
                String s = part;
                if (s == null || s.isEmpty()) continue;
                if (s.startsWith("data:")) {
                    s = s.substring(5);
                    if (s.startsWith(" ")) s = s.substring(1);
                }
                if ("[DONE]".equals(s.trim())) {
                    doneSeen.set(true);
                    continue;
                }
                try {
                    JSONObject obj = JSON.parseObject(s);
                    JSONArray choices = obj.getJSONArray("choices");
                    if (choices != null && !choices.isEmpty()) {
                        JSONObject c0 = choices.getJSONObject(0);
                        JSONObject delta = c0.getJSONObject("delta");
                        if (delta != null) {
                            String content = delta.getString("content");
                            if (content != null) {
                                try { queue.offer(content); } catch (Exception ignore) {}
                            }
                        }
                    }
                } catch (Exception e) {
                    e.printStackTrace();
                }
            }
        }, err -> {
            errorSeen.set(true);
            doneSeen.set(true);
            if (onError != null) onError.accept(err);
        }, () -> {
            if (!doneSeen.get()) {
                errorSeen.set(true);
                doneSeen.set(true);
                if (onError != null) onError.accept(new RuntimeException("LLM 流意外完成"));
            } else {
                doneSeen.set(true);
            }
        });
    }
}