添加对通用文本向量的支持

README-zh.md

@@ -6,7 +6,7 @@
 
 ##功能特性
 
-**支持阿里云服务的通义千问和通义万相**
+**支持阿里云的（DashScope）灵积服务通义千问和通义万相**
 
 提示：如果要本地通义千问模型可以使用官方的Ollama模块进行
 

models/spring-ai-dashscope/src/main/java/org/springframework/ai/dashcope/DashCopeService.java

@@ -132,14 +132,15 @@ public record ChatCompletionRequest(
 			@JsonProperty("parameters") Parameters parameters) {
 
 		public ChatCompletionRequest(List<ChatCompletionMessage> messages,String model,Float temperature) {
-			this(model,new Input(null, messages),new Parameters("message", null, null, null, null, null, temperature, null, null));
+			this(model,new Input(null, messages,null),new Parameters("message", null, null, null, null, null, temperature, null, null,null));
 		}
 	}
 
 	@JsonInclude(Include.NON_NULL)
 	public record Input(
 			@JsonProperty("prompt") String prompt,
-			@JsonProperty("messages") List<ChatCompletionMessage> messages) {
+			@JsonProperty("messages") List<ChatCompletionMessage> messages,
+			@JsonProperty("texts")List<String> texts) {
 
 	}
 
@@ -160,7 +161,12 @@ public record Parameters(
 			@JsonProperty("repetition_penalty")Float repetitionPenlty,
 			@JsonProperty("temperature") Float temperature,
 			@JsonProperty("stop")List<String> stop,
-			@JsonProperty("incremental_output")Boolean incrementalOutput) {	
+			@JsonProperty("incremental_output")Boolean incrementalOutput,
+			@JsonProperty("text_type")String textType) {	
+
+		public Parameters(String textType) {
+			this(null,null,null,null,null,null,null,null, null, textType);
+		}
 	}
 
 	@JsonInclude(Include.NON_NULL)
@@ -185,7 +191,14 @@ public record Output(
 			@JsonProperty("task_id") String taskId,
 			@JsonProperty("task_status") StatusStatus taskStatus,
 			@JsonProperty("task_metrics") TaskMetrices taskMetrices,
-			@JsonProperty("results") List<Results> results) {
+			@JsonProperty("results") List<Results> results,
+			@JsonProperty("embeddings") List<Embedding> embeddings) {
+
+	}
+
+	public record Embedding(
+			@JsonProperty("text_index")Integer textIndex,
+			@JsonProperty("embedding") List<Double> embedding) {
 
 	}
 
@@ -235,6 +248,20 @@ public record QWenImageResponse(
 
 	}
 
+	@JsonInclude(Include.NON_NULL)
+	public record EmbeddingRequest(
+			String model,Input input,Parameters parameters) {
+
+		public EmbeddingRequest(String model,List<String> texts,String textType) {
+			this(model, new Input(null, null,texts), new Parameters(textType));
+		}
+	}
+
+	public record EmbeddingResponse(
+			String requestId,Usage usage,Output output) {
+
+	}
+
 	public enum StatusStatus{
 		PENDING,RUNNING,SUCCEEDED,FAILED,UNKNOWN,
 	}
@@ -265,6 +292,13 @@ public ResponseEntity<QWenImageResponse> findImageTaskResult(String taskId){
 		return this.restClient.get().uri("/api/v1/tasks/{task_id}",taskId).retrieve().toEntity(QWenImageResponse.class);
 	}
 
+	public ResponseEntity<EmbeddingResponse> embeddingRequest(EmbeddingRequest embeddingRequest) {
+		return this.restClient.post()
+				.uri("/api/v1/services/embeddings/text-embedding/text-embedding")
+				.body(embeddingRequest).retrieve()
+				.toEntity(EmbeddingResponse.class);
+	}
+
 	public static void main(String[] args) throws JsonProcessingException {
 		String body = "{\"output\":{\"finish_reason\":\"stop\",\"text\":\"我是阿里云开发的一款超大规模语言模型，我叫通义千问。\"},\"usage\":{\"total_tokens\":20,\"output_tokens\":17,\"input_tokens\":3},\"request_id\":\"f978f627-fd0f-91fd-be5d-b3ec1ac394b1\"}";
 		ObjectMapper objectMapper = new ObjectMapper();

models/spring-ai-dashscope/src/main/java/org/springframework/ai/dashcope/qwen/QWenEmbeddingClient.java

@@ -1,14 +1,26 @@
 package org.springframework.ai.dashcope.qwen;
 
+import static org.assertj.core.api.Assertions.contentOf;
+
+import java.time.Duration;
 import java.util.List;
 
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
 import org.springframework.ai.dashcope.DashCopeService;
+import org.springframework.ai.dashcope.DashCopeService.DashCopeApiException;
+import org.springframework.ai.dashcope.DashCopeService.Usage;
 import org.springframework.ai.dashcope.metadata.support.EmbeddingModel;
 import org.springframework.ai.document.Document;
 import org.springframework.ai.document.MetadataMode;
 import org.springframework.ai.embedding.AbstractEmbeddingClient;
+import org.springframework.ai.embedding.Embedding;
 import org.springframework.ai.embedding.EmbeddingRequest;
 import org.springframework.ai.embedding.EmbeddingResponse;
+import org.springframework.ai.embedding.EmbeddingResponseMetadata;
+import org.springframework.http.ResponseEntity;
+import org.springframework.retry.RetryListener;
+import org.springframework.retry.support.RetryTemplate;
 import org.springframework.util.Assert;
 
 /**
@@ -17,8 +29,19 @@
  */
 public class QWenEmbeddingClient extends AbstractEmbeddingClient {
 
+	private static final Logger logger = LoggerFactory.getLogger(QWenEmbeddingClient.class);
+
 	public static final EmbeddingModel DEFAULT_OPENAI_EMBEDDING_MODEL = EmbeddingModel.TEXT_EMBEDDING_V1;
 
+	private final RetryTemplate retryTemplate = RetryTemplate.builder().maxAttempts(10)
+			.retryOn(DashCopeApiException.class)
+			.exponentialBackoff(Duration.ofMillis(2000), 5, Duration.ofMillis(3* 60000))
+			.withListener(new RetryListener() {
+				public <T extends Object, E extends Throwable> void onError(org.springframework.retry.RetryContext context, org.springframework.retry.RetryCallback<T,E> callback, Throwable throwable) {
+					logger.warn("Retry error. Retry count:"+context.getRetryCount(),throwable);
+				};
+			}).build();
+
 	private final QWenEmbeddingOptions embeddingOptions;
 
 	private final DashCopeService dashCopeService;
@@ -51,8 +74,23 @@ public List<Double> embed(Document document) {
 
 	@Override
 	public EmbeddingResponse call(EmbeddingRequest request) {
-		// TODO Auto-generated method stub
-		return null;
+		return this.retryTemplate.execute(ctx -> {
+			org.springframework.ai.dashcope.DashCopeService.EmbeddingRequest embeddingRequest = new org.springframework.ai.dashcope.DashCopeService.EmbeddingRequest(this.embeddingOptions.getModel().value, request.getInstructions(), "document");
+			org.springframework.ai.dashcope.DashCopeService.EmbeddingResponse embeddingResponse = this.dashCopeService.embeddingRequest(embeddingRequest).getBody();
+			if(embeddingResponse == null) {
+				logger.warn("No {}",embeddingRequest);
+			}
+			var metadata = generateResponseMetadata(embeddingRequest.model(),embeddingResponse.usage());
+			List<Embedding> embeddings = embeddingResponse.output().embeddings().stream().map(e-> new Embedding(e.embedding(),e.textIndex())).toList();
+			return new EmbeddingResponse(embeddings,metadata);
+		});
+	}
+
+	private EmbeddingResponseMetadata generateResponseMetadata(String model,Usage usage) {
+		EmbeddingResponseMetadata metadata = new EmbeddingResponseMetadata();
+		metadata.put("model", model);
+		metadata.put("total-tokens", usage.totalTokens());
+		return metadata;
 	}
 
 }

models/spring-ai-dashscope/src/main/java/org/springframework/ai/dashcope/qwen/QWenEmbeddingOptions.java

@@ -33,5 +33,9 @@ public QWenEmbeddingOptions build() {
 			return this.embeddingOptions;
 		}
 	}
+
+	public EmbeddingModel getModel() {
+		return model;
+	}
 
 }

models/spring-ai-dashscope/src/main/java/org/springframework/ai/dashcope/qwen/QWenImageClient.java

@@ -52,7 +52,7 @@ public ImageResponse call(ImagePrompt imagePrompt) {
 
 			String instructions = imagePrompt.getInstructions().get(0).getText();
 
-			QWenImageRequest imageRequest = new QWenImageRequest(new Input(instructions,null), null);
+			QWenImageRequest imageRequest = new QWenImageRequest(new Input(instructions,null,null), null);
 			ResponseEntity<QWenImageResponse> responseEntity = this.dashCopeService.createQwenImageTask(imageRequest);
 			return responseEntity.getBody();
 		});

models/spring-ai-dashscope/src/test/java/org/springframework/ai/dashcope/embedding/EmbeddingIT.java

@@ -5,6 +5,9 @@
 import java.util.List;
 
 import org.junit.jupiter.api.Test;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+import org.springframework.ai.dashcope.chat.QwenChatClientIT;
 import org.springframework.ai.dashcope.qwen.QWenEmbeddingClient;
 import org.springframework.ai.embedding.EmbeddingResponse;
 import org.springframework.beans.factory.annotation.Autowired;
@@ -13,6 +16,8 @@
 @SpringBootTest
 public class EmbeddingIT {
 
+	private final Logger logger = LoggerFactory.getLogger(EmbeddingIT.class);
+
 	@Autowired
 	private QWenEmbeddingClient embeddingClient;
 
@@ -22,6 +27,7 @@ public void simpleEmbedding() {
 
 		EmbeddingResponse embeddingResponse = embeddingClient.embedForResponse(List.of("Hello word"));
 		assertThat(embeddingResponse.getResults()).hasSize(1);
+		logger.info("{}",embeddingResponse.getResult().getOutput());
 		assertThat(embeddingResponse.getResults().get(0)).isNotNull();
 		//assertThat(embeddingResponse.getMetadata()).containsEntry("model", "");
 	}