topdog/supersonic
1
0
Fork 0
mirror of https://github.com/tencentmusic/supersonic.git synced 2025-12-11 03:58:14 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

(improvement)(common|headless|chat|auth) 鉴权优化与召回优化

Browse Source 
1 修复生成的用户token 一生成就失效的问题
2 如果用户设置的token ，需校验是否数据库存在，因为用户可设置一年的token 有泄露风险
3 结果解析优化， 去除不可以解析的情况，解析问题需要改写后的问，
4 召回样例，用相似度，保住至少有一个样例是高相似度的
5 数据集召回，填加完全匹配格式筛选逻辑
This commit is contained in:
guilinlewis
2025-06-23 10:03:17 +08:00
parent 7e6639df83
commit 1faf84e372
10 changed files with 36 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
headless/api/src/main/java/com/tencent/supersonic/headless/api/pojo/SemanticParseInfo.java
View File

@@ -66,7 +66,7 @@ public class SemanticParseInfo implements Serializable {
DataSetMatchResult mr2 = getDataSetMatchResult(o2.getElementMatches());
double difference = mr1.getMaxDatesetSimilarity() - mr2.getMaxDatesetSimilarity();
if (Math.abs(difference) < 0.0005) { // 看完全匹配的个数，实践证明，可以用户输入规范后，该逻辑具有优势
if (Math.abs(difference) < 0.0005) { // 看完全匹配的个数，实践证明，可以用户输入规范后，该逻辑具有优势
if (!o1.getDataSetId().equals(o2.getDataSetId())) {
List<SchemaElementMatch> elementMatches1 = o1.getElementMatches().stream()
.filter(e -> e.getSimilarity() == 1).collect(Collectors.toList());

2
headless/chat/src/main/java/com/tencent/supersonic/headless/chat/parser/llm/PromptHelper.java
View File

@@ -49,7 +49,7 @@ public class PromptHelper {
// use random collection of exemplars for each self-consistency inference
for (int i = 0; i < selfConsistencyNumber; i++) {
List<Text2SQLExemplar> shuffledList = new ArrayList<>(exemplars);
List<Text2SQLExemplar> same = shuffledList.stream() // 相似度极高的话，先找出来
List<Text2SQLExemplar> same = shuffledList.stream() // 相似度极高的话，先找出来
.filter(e -> e.getSimilarity() > 0.989).collect(Collectors.toList());
List<Text2SQLExemplar> noSame = shuffledList.stream()
.filter(e -> e.getSimilarity() <= 0.989).collect(Collectors.toList());