Add FIM (fill-in-the-middle) completion support for DeepSeek

auto-configurations/models/spring-ai-autoconfigure-model-deepseek/src/main/java/org/springframework/ai/model/deepseek/autoconfigure/DeepSeekChatProperties.java

@@ -76,6 +76,10 @@ public class DeepSeekChatProperties extends DeepSeekParentProperties {
 
 	private @Nullable ReasoningEffort reasoningEffort;
 
+	private @Nullable Boolean echo;
+
+	private @Nullable String suffix;
+
 	public boolean isEnabled() {
 		return this.enabled;
 	}
@@ -196,6 +200,22 @@ public void setReasoningEffort(@Nullable ReasoningEffort reasoningEffort) {
 		this.reasoningEffort = reasoningEffort;
 	}
 
+	public @Nullable Boolean getEcho() {
+		return this.echo;
+	}
+
+	public void setEcho(@Nullable Boolean echo) {
+		this.echo = echo;
+	}
+
+	public @Nullable String getSuffix() {
+		return this.suffix;
+	}
+
+	public void setSuffix(@Nullable String suffix) {
+		this.suffix = suffix;
+	}
+
 	public DeepSeekChatOptions toOptions() {
 		return DeepSeekChatOptions.builder()
 			.model(this.model)
@@ -210,6 +230,8 @@ public DeepSeekChatOptions toOptions() {
 			.topLogprobs(this.topLogprobs)
 			.thinking(this.thinking)
 			.reasoningEffort(this.reasoningEffort)
+			.echo(this.echo)
+			.suffix(this.suffix)
 			.build();
 	}
 
@@ -347,6 +369,26 @@ public void setReasoningEffort(@Nullable ReasoningEffort reasoningEffort) {
 			DeepSeekChatProperties.this.setReasoningEffort(reasoningEffort);
 		}
 
+		@DeprecatedConfigurationProperty(replacement = "spring.ai.deepseek.chat.echo")
+		@Deprecated(since = "2.0.0", forRemoval = true)
+		public @Nullable Boolean getEcho() {
+			return DeepSeekChatProperties.this.getEcho();
+		}
+
+		public void setEcho(@Nullable Boolean echo) {
+			DeepSeekChatProperties.this.setEcho(echo);
+		}
+
+		@DeprecatedConfigurationProperty(replacement = "spring.ai.deepseek.chat.suffix")
+		@Deprecated(since = "2.0.0", forRemoval = true)
+		public @Nullable String getSuffix() {
+			return DeepSeekChatProperties.this.getSuffix();
+		}
+
+		public void setSuffix(@Nullable String suffix) {
+			DeepSeekChatProperties.this.setSuffix(suffix);
+		}
+
 	}
 
 }

auto-configurations/models/spring-ai-autoconfigure-model-deepseek/src/test/java/org/springframework/ai/model/deepseek/autoconfigure/DeepSeekPropertiesTests.java

@@ -119,7 +119,9 @@ public void chatOptionsTest() {
 				"spring.ai.deepseek.chat.top-p=0.56",
 				"spring.ai.deepseek.chat.user=userXYZ",
 				"spring.ai.deepseek.chat.thinking.type=disabled",
-				"spring.ai.deepseek.chat.reasoning-effort=max"
+				"spring.ai.deepseek.chat.reasoning-effort=max",
+				"spring.ai.deepseek.chat.echo=true",
+				"spring.ai.deepseek.chat.suffix=return x"
 				)
 			// @formatter:on
 			.withConfiguration(AutoConfigurations.of(DeepSeekChatAutoConfiguration.class,
@@ -141,6 +143,13 @@ public void chatOptionsTest() {
 				assertThat(chatProperties.getTopP()).isEqualTo(0.56);
 				assertThat(chatProperties.getThinking()).isEqualTo(Thinking.DISABLED);
 				assertThat(chatProperties.getReasoningEffort()).isEqualTo(ReasoningEffort.MAX);
+
+				assertThat(chatProperties.getEcho()).isTrue();
+				assertThat(chatProperties.getSuffix()).isEqualTo("return x");
+
+				var chatOptions = chatProperties.toOptions();
+				assertThat(chatOptions.getEcho()).isTrue();
+				assertThat(chatOptions.getSuffix()).isEqualTo("return x");
 			});
 	}
 

models/spring-ai-deepseek/src/main/java/org/springframework/ai/deepseek/DeepSeekChatModel.java

@@ -437,6 +437,12 @@ else if (message.getMessageType() == MessageType.TOOL) {
 		if (options.getReasoningEffort() != null) {
 			requestBuilder.reasoningEffort(options.getReasoningEffort());
 		}
+		if (options.getEcho() != null) {
+			requestBuilder.echo(options.getEcho());
+		}
+		if (options.getSuffix() != null) {
+			requestBuilder.suffix(options.getSuffix());
+		}
 
 		// Add the tool definitions to the request's tools parameter.
 		List<ToolDefinition> toolDefinitions = this.toolCallingManager.resolveToolDefinitions(options);

models/spring-ai-deepseek/src/main/java/org/springframework/ai/deepseek/DeepSeekChatOptions.java

@@ -142,6 +142,16 @@ public class DeepSeekChatOptions implements ToolCallingChatOptions {
 	 */
 	private final @Nullable ReasoningEffort reasoningEffort;
 
+	/**
+	 * Echo back the prompt in addition to the completion.
+	 */
+	private final @Nullable Boolean echo;
+
+	/**
+	 * The suffix that comes after a completion of inserted text.
+	 */
+	private final @Nullable String suffix;
+
 	/**
 	 * Tool Function Callbacks to register with the ChatModel.
 	 * For Prompt Options the toolCallbacks are automatically enabled for the duration of the prompt execution.
@@ -158,7 +168,8 @@ protected DeepSeekChatOptions(@Nullable String model, @Nullable Double frequency
 			@Nullable List<String> stop, @Nullable Double temperature, @Nullable Double topP,
 			@Nullable Boolean logprobs, @Nullable Integer topLogprobs, @Nullable List<DeepSeekApi.FunctionTool> tools,
 			@Nullable Object toolChoice, @Nullable Thinking thinking, @Nullable ReasoningEffort reasoningEffort,
-			@Nullable List<ToolCallback> toolCallbacks, @Nullable Map<String, Object> toolContext) {
+			@Nullable Boolean echo, @Nullable String suffix, @Nullable List<ToolCallback> toolCallbacks,
+			@Nullable Map<String, Object> toolContext) {
 		this.model = model != null ? model : DeepSeekApi.DEFAULT_CHAT_MODEL.getValue();
 		this.frequencyPenalty = frequencyPenalty;
 		this.maxTokens = maxTokens;
@@ -173,6 +184,8 @@ protected DeepSeekChatOptions(@Nullable String model, @Nullable Double frequency
 		this.toolChoice = toolChoice;
 		this.thinking = thinking;
 		this.reasoningEffort = reasoningEffort;
+		this.echo = echo;
+		this.suffix = suffix;
 		this.toolCallbacks = toolCallbacks != null ? List.copyOf(toolCallbacks) : null;
 		this.toolContext = toolContext != null ? Map.copyOf(toolContext) : null;
 	}
@@ -253,6 +266,14 @@ public String getModel() {
 		return this.topLogprobs;
 	}
 
+	public @Nullable Boolean getEcho() {
+		return this.echo;
+	}
+
+	public @Nullable String getSuffix() {
+		return this.suffix;
+	}
+
 	@Override
 	public @Nullable Integer getTopK() {
 		return null;
@@ -285,14 +306,17 @@ public Builder mutate() {
 			.tools(this.tools)
 			.toolChoice(this.toolChoice)
 			.thinking(this.thinking)
-			.reasoningEffort(this.reasoningEffort);
+			.reasoningEffort(this.reasoningEffort)
+			.echo(this.echo)
+			.suffix(this.suffix);
 	}
 
 	@Override
 	public int hashCode() {
 		return Objects.hash(this.model, this.frequencyPenalty, this.logprobs, this.topLogprobs, this.maxTokens,
 				this.presencePenalty, this.responseFormat, this.stop, this.temperature, this.topP, this.tools,
-				this.toolChoice, this.thinking, this.reasoningEffort, this.toolCallbacks, this.toolContext);
+				this.toolChoice, this.thinking, this.reasoningEffort, this.echo, this.suffix, this.toolCallbacks,
+				this.toolContext);
 	}
 
 	@Override
@@ -312,8 +336,8 @@ public boolean equals(@Nullable Object o) {
 				&& Objects.equals(this.temperature, other.temperature) && Objects.equals(this.topP, other.topP)
 				&& Objects.equals(this.tools, other.tools) && Objects.equals(this.toolChoice, other.toolChoice)
 				&& Objects.equals(this.thinking, other.thinking)
-				&& Objects.equals(this.reasoningEffort, other.reasoningEffort)
-				&& Objects.equals(this.toolCallbacks, other.toolCallbacks)
+				&& Objects.equals(this.reasoningEffort, other.reasoningEffort) && Objects.equals(this.echo, other.echo)
+				&& Objects.equals(this.suffix, other.suffix) && Objects.equals(this.toolCallbacks, other.toolCallbacks)
 				&& Objects.equals(this.toolContext, other.toolContext);
 	}
 
@@ -347,6 +371,10 @@ public B clone() {
 
 		protected @Nullable ReasoningEffort reasoningEffort;
 
+		protected @Nullable Boolean echo;
+
+		protected @Nullable String suffix;
+
 		public B model(DeepSeekApi.@Nullable ChatModel deepseekAiChatModel) {
 			if (deepseekAiChatModel == null) {
 				this.model = null;
@@ -416,6 +444,16 @@ public B reasoningEffortMax() {
 			return self();
 		}
 
+		public B echo(@Nullable Boolean echo) {
+			this.echo = echo;
+			return self();
+		}
+
+		public B suffix(@Nullable String suffix) {
+			this.suffix = suffix;
+			return self();
+		}
+
 		public B combineWith(ChatOptions.Builder<?> other) {
 			super.combineWith(other);
 			if (other instanceof AbstractBuilder<?> that) {
@@ -447,6 +485,12 @@ public B combineWith(ChatOptions.Builder<?> other) {
 				if (that.reasoningEffort != null) {
 					this.reasoningEffort = that.reasoningEffort;
 				}
+				if (that.echo != null) {
+					this.echo = that.echo;
+				}
+				if (that.suffix != null) {
+					this.suffix = that.suffix;
+				}
 			}
 			return self();
 		}
@@ -455,8 +499,8 @@ public B combineWith(ChatOptions.Builder<?> other) {
 		public DeepSeekChatOptions build() {
 			return new DeepSeekChatOptions(this.model, this.frequencyPenalty, this.maxTokens, this.presencePenalty,
 					this.responseFormat, this.stopSequences, this.temperature, this.topP, this.logprobs,
-					this.topLogprobs, this.tools, this.toolChoice, this.thinking, this.reasoningEffort,
-					this.toolCallbacks, this.toolContext);
+					this.topLogprobs, this.tools, this.toolChoice, this.thinking, this.reasoningEffort, this.echo,
+					this.suffix, this.toolCallbacks, this.toolContext);
 		}
 
 	}

models/spring-ai-deepseek/src/main/java/org/springframework/ai/deepseek/api/DeepSeekApi.java

@@ -42,6 +42,7 @@
 import org.springframework.http.MediaType;
 import org.springframework.http.ResponseEntity;
 import org.springframework.util.Assert;
+import org.springframework.util.StringUtils;
 import org.springframework.web.client.ResponseErrorHandler;
 import org.springframework.web.client.RestClient;
 import org.springframework.web.reactive.function.client.WebClient;
@@ -218,7 +219,7 @@ private String getEndpoint(ChatCompletionRequest request) {
 			.map(ChatCompletionMessage::prefix)
 			.filter(Objects::nonNull)
 			.anyMatch(prefix -> prefix);
-		String endpointPrefix = isPrefix ? this.betaPrefixPath : "";
+		String endpointPrefix = isPrefix || StringUtils.hasText(request.suffix) ? this.betaPrefixPath : "";
 		return endpointPrefix + this.completionsPath;
 	}
 
@@ -517,7 +518,9 @@ public record ChatCompletionRequest(// @formatter:off
 			@JsonProperty("tools") @Nullable List<FunctionTool> tools,
 			@JsonProperty("tool_choice") @Nullable Object toolChoice,
 			@JsonProperty("thinking") @Nullable Thinking thinking,
-			@JsonProperty("reasoning_effort") @Nullable ReasoningEffort reasoningEffort) {
+			@JsonProperty("reasoning_effort") @Nullable ReasoningEffort reasoningEffort,
+			@JsonProperty("echo") @Nullable Boolean echo,
+			@JsonProperty("suffix") @Nullable String suffix) {
 
 		/**
 		 * Create a new {@link ChatCompletionRequest} builder.
@@ -537,7 +540,7 @@ public static Builder builder() {
 		public ChatCompletionRequest(List<ChatCompletionMessage> messages, Boolean stream) {
 			this(messages, null, null, null, null, null,
 					null, stream, null, null, null, null, null,
-					null, null, null);
+					null, null, null, null, null);
 		}
 
 		/**
@@ -550,7 +553,7 @@ public ChatCompletionRequest(List<ChatCompletionMessage> messages, Boolean strea
 		public ChatCompletionRequest(List<ChatCompletionMessage> messages, String model, Double temperature) {
 			this(messages, model, null,
 		null, null, null, null,  false,  temperature, null,
-			null, null, null, null, null, null);
+			null, null, null, null, null, null, null, null);
 		}
 
 		/**
@@ -565,7 +568,7 @@ public ChatCompletionRequest(List<ChatCompletionMessage> messages, String model,
 		public ChatCompletionRequest(List<ChatCompletionMessage> messages, String model, Double temperature, boolean stream) {
 			this(messages, model, null,
 					null, null, null, null,  stream,  temperature, null,
-				null, null, null, null, null, null);
+				null, null, null, null, null, null, null, null);
 		}
 
 		/**
@@ -683,6 +686,10 @@ public static class Builder {
 
 			private @Nullable Thinking thinking;
 
+			private @Nullable Boolean echo;
+
+			private @Nullable String suffix;
+
 			/**
 			 * Set the messages comprising the conversation so far.
 			 * @param messages the messages.
@@ -843,6 +850,26 @@ public Builder thinking(@Nullable Thinking thinking) {
 				return this;
 			}
 
+			/**
+			 * Set whether to echo the prompt.
+			 * @param echo whether to echo the prompt.
+			 * @return this builder.
+			 */
+			public Builder echo(@Nullable Boolean echo) {
+				this.echo = echo;
+				return this;
+			}
+
+			/**
+			 * Set the suffix.
+			 * @param suffix the suffix.
+			 * @return this builder.
+			 */
+			public Builder suffix(@Nullable String suffix) {
+				this.suffix = suffix;
+				return this;
+			}
+
 			/**
 			 * Build the {@link ChatCompletionRequest}.
 			 * @return a new {@link ChatCompletionRequest}.
@@ -852,7 +879,7 @@ public ChatCompletionRequest build() {
 				return new ChatCompletionRequest(this.messages, this.model, this.frequencyPenalty, this.maxTokens,
 						this.presencePenalty, this.responseFormat, this.stop, this.stream, this.temperature, this.topP,
 						this.logprobs, this.topLogprobs, this.tools, this.toolChoice, this.thinking,
-						this.reasoningEffort);
+						this.reasoningEffort, this.echo, this.suffix);
 			}
 
 		}