DATA_MIDDLE_PLATFORM/Data_Middle_Platform_Server
3
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

refactor(yudao-module-llm): 调整模型补全请求的默认参数并优化数据处理

Browse Source 
- 将 max_tokens 参数从 4000调整为 4096
- 将 temperature 参数从0.7 调整为 0.2
- 新增 top_p 参数，默认值为0.9
- 优化了数据处理逻辑，移除了不必要的注释代码
- 调整了 SseEmitter 的心跳检测机制
- 优化了数据解析过程，移除了多余的换行符
This commit is contained in:
Liuyang 2025-03-03 18:38:00 +08:00
parent e0851f472a
commit 0340b24f78
2 changed files with 43 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

43
yudao-module-llm/yudao-module-llm-biz/src/main/java/cn/iocoder/yudao/module/llm/service/http/ModelService.java
View File

@ -148,6 +148,9 @@ public class ModelService {
*/
public ModelCompletionsRespVO modelCompletionsStream (String url, ModelCompletionsReqVO req, SseEmitter emitter, String uuid) {
req.setStream(true);
req.setTemperature(0.2);
req.setTop_p(0.9);
req.setMax_tokens(4096);
log.info("开始处理模型补全请求");
// 检查模型是否为空，若为空则设置默认模型
@ -175,6 +178,38 @@ public class ModelService {
}
//
// // 解析响应内容
// log.info("正在解析响应内容...");
// ChatCompletion chatCompletion = JSON.parseObject(result, ChatCompletion.class);
//
// // 检查响应内容是否包含错误信息
// if (StringUtils.isBlank(chatCompletion.getDetail())) {
// log.info("响应内容无错误信息，提取回答内容...");
//
// // 提取回答内容
// String respContent = chatCompletion.getChoices().get(0).getMessage().getContent();
// String patternString = "(<think>.*?</think>)";
// Pattern pattern = Pattern.compile(patternString, Pattern.DOTALL);
// Matcher matcher = pattern.matcher(respContent);
// String answerContent = matcher.replaceAll("");
//
// // 构建返回对象
// ModelCompletionsRespVO respVO = new ModelCompletionsRespVO();
// respVO.setSystem("助手");
// respVO.setQuestion(req.getMessages().get(req.getMessages().size() - 1).getContent());
// respVO.setAnswer(answer);
//
// log.info("模型补全请求处理成功。返回结果: {}", JSON.toJSONString(respVO));
// return respVO;
//
// } else {
// log.warn("响应内容包含错误信息，返回 null");
// return null;
// }
// return null;
// 构建返回对象
ModelCompletionsRespVO respVO = new ModelCompletionsRespVO();
respVO.setSystem("助手");
respVO.setQuestion(req.getMessages().get(req.getMessages().size() - 1).getContent());
@ -249,6 +284,7 @@ public class ModelService {
}
log.info("接收到的响应行数据: {}", line);
line = line.replaceAll("\n", " ");
String content = parseStreamLine(line, uuid);
if (content != null) {
emitter.send(
@ -263,8 +299,9 @@ public class ModelService {
}
// // 心跳检测
// if (System.currentTimeMillis() - lastSendTime > MAX_IDLE_TIME && !result.toString().isEmpty()) {
// emitter.send("[HEARTBEAT]");
// if (System.currentTimeMillis() - lastSendTime > 15_000) {
// emitter.send(SseEmitter.event().comment("heartbeat"));
// lastSendTime = System.currentTimeMillis();
// }
}
emitter.complete();
@ -285,7 +322,7 @@ public class ModelService {
if (StringUtils.isNotBlank(line)) {
if (line.startsWith("data: ")) {
String dataString = extractJsonFromDataString(line);
if (!dataString.contains("[DONE]")||!dataString.contains("</think>")) {
if (!dataString.contains("[DONE]")) {
JSONObject jsonObject = JSON.parseObject(dataString);
// 获取 choices 数组
JSONArray choicesArray = jsonObject.getJSONArray("choices");

8
yudao-module-llm/yudao-module-llm-biz/src/main/java/cn/iocoder/yudao/module/llm/service/http/vo/ModelCompletionsReqVO.java
View File

@ -15,12 +15,10 @@ public class ModelCompletionsReqVO {
private String model;
private List<ModelCompletionsMessage> messages;
private Integer max_tokens = 4000;
private Double temperature = 0.7;
// private Integer max_tokens;
// private Double temperature;
private Integer max_tokens = 4096;
private Double temperature = 0.2;
private Double top_p = 0.9;
private Boolean stream;
// private Integer max_length = 120000;
@Data
@NoArgsConstructor