topdog/supersonic
1
0
Fork 0
mirror of https://github.com/tencentmusic/supersonic.git synced 2025-12-11 20:25:12 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

(improvement)(chat) Reduce frequent loading of embedding models to improve loading performance. (#1478)

Browse Source
This commit is contained in:
lexluo09
2024-07-30 11:25:03 +08:00
committed by GitHub
parent 23af977972
commit 9a1fac5d4c
6 changed files with 55 additions and 40 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
common/src/main/java/dev/langchain4j/inmemory/spring/InMemoryAutoConfig.java
View File

@@ -5,6 +5,7 @@ import dev.langchain4j.model.embedding.AllMiniLmL6V2QuantizedEmbeddingModel;
import dev.langchain4j.model.embedding.BgeSmallZhEmbeddingModel;
import dev.langchain4j.model.embedding.EmbeddingModel;
import dev.langchain4j.model.embedding.S2OnnxEmbeddingModel;
import dev.langchain4j.provider.EmbeddingModelConstant;
import dev.langchain4j.store.embedding.EmbeddingStoreFactory;
import org.apache.commons.lang3.StringUtils;
import org.springframework.boot.autoconfigure.condition.ConditionalOnProperty;
@@ -17,10 +18,6 @@ import static dev.langchain4j.inmemory.spring.Properties.PREFIX;
@Configuration
@EnableConfigurationProperties(Properties.class)
public class InMemoryAutoConfig {
public static final String BGE_SMALL_ZH = "bge-small-zh";
public static final String ALL_MINILM_L6_V2 = "all-minilm-l6-v2-q";
@Bean
@ConditionalOnProperty(PREFIX + ".embedding-store.persist-path")
EmbeddingStoreFactory inMemoryChatModel(Properties properties) {
@@ -37,10 +34,10 @@ public class InMemoryAutoConfig {
return new S2OnnxEmbeddingModel(modelPath, vocabularyPath);
}
String modelName = embeddingModelProperties.getModelName();
if (BGE_SMALL_ZH.equalsIgnoreCase(modelName)) {
if (EmbeddingModelConstant.BGE_SMALL_ZH.equalsIgnoreCase(modelName)) {
return new BgeSmallZhEmbeddingModel();
}
if (ALL_MINILM_L6_V2.equalsIgnoreCase(modelName)) {
if (EmbeddingModelConstant.ALL_MINILM_L6_V2.equalsIgnoreCase(modelName)) {
return new AllMiniLmL6V2QuantizedEmbeddingModel();
}
return new BgeSmallZhEmbeddingModel();