Springai 指定模型的三种方式（Ollama）
在实际开发中，Ollama 支持三种常用的模型指定方式：
1. 从 yml 配置读取默认模型
注意： 这是最基础、最推荐的方式，必须先配置好才能用自动注入的 OllamaChatModel。
spring:ai:ollama:base-url: http://localhost:11434chat:options:model: deepseek-r1:7b
@Autowired
private OllamaChatModel chatModel;
// 直接调用 chatModel.call(...) 即用默认模型
2. Prompt 临时指定模型
通过 Prompt 构造时传入 OllamaOptions，可临时切换模型：
import org.springframework.ai.ollama.api.OllamaOptions;Prompt prompt = new Prompt(messageList,OllamaOptions.builder().model("qwen2.5-vl") // 临时指定模型.build());return chatModel.stream(prompt);
3. 创建多个 OllamaChatModel 动态切换
可在配置类中为不同模型创建多个 Bean，或用工厂模式动态切换：
@Bean
public OllamaChatModel ollamaQwenModel() {OllamaApi ollamaApi = OllamaApi.builder().baseUrl("http://localhost:11434").build();// 解析参数OllamaOptions.Builder optionsBuilder = OllamaOptions.builder().model("qwen2.5-vl:3b");return OllamaChatModel.builder().ollamaApi(ollamaApi).defaultOptions(optionsBuilder.build()).build();
}@Bean
public OllamaChatModel ollamaLlamaModel() {OllamaApi ollamaApi = OllamaApi.builder().baseUrl("http://localhost:11434").build();// 解析参数OllamaOptions.Builder optionsBuilder = OllamaOptions.builder().model("llama2:7b");return OllamaChatModel.builder().ollamaApi(ollamaApi).defaultOptions(optionsBuilder.build()).build();
}
或通过自定义工厂类，根据参数动态返回不同模型实例：
public class DynamicModelFactory {public OllamaChatModel getModelByName(String modelName) {// ...根据modelName返回不同OllamaChatModel实例...}
}
接口调用时根据参数动态切换：
OllamaChatModel model = dynamicModelFactory.getModelByName(modelName);
return model.stream(prompt);
建议所有模型还是维护到数据库，因为大部分模型特别是相同供应商的调用方式都一样
CREATE TABLE `ai_model` (`id` bigint NOT NULL AUTO_INCREMENT,`vendor` varchar(64) NOT NULL COMMENT '供应商',`icon` varchar(255) DEFAULT NULL COMMENT '图标URL',`name` varchar(128) NOT NULL COMMENT '模型名称',`api_key` varchar(255) DEFAULT NULL COMMENT '密钥',`api_url` varchar(255) NOT NULL COMMENT '模型API地址',`tags` varchar(255) DEFAULT NULL COMMENT '标签（推理、对话、图片、语音等，逗号分隔）',`type` varchar(32) NOT NULL COMMENT '类型（对话、图片、音频、视频、量化）',`status` tinyint NOT NULL DEFAULT 1 COMMENT '模型可用状态 1:可用 0:不可用',`description` text COMMENT '模型描述',`params` json DEFAULT NULL COMMENT '模型参数（如温度等）',`create_time` timestamp NULL DEFAULT CURRENT_TIMESTAMP,`update_time` timestamp NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,PRIMARY KEY (`id`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_german2_ci;
DynamicModelFactory.java
/*** 动态模型工厂，项目启动时缓存所有模型*/
@Component
public class DynamicModelFactory {@Autowiredprivate AiModelRepository aiModelRepository;private final HashMap<String, MyModel> modelHashMap = new HashMap<>();@PostConstructpublic void init() {List<AiModel> models = aiModelRepository.findAll();ObjectMapper objectMapper = new ObjectMapper();for (AiModel m : models) {MyModel myModel = new MyModel();myModel.vendor = m.getVendor();myModel.name = m.getName();myModel.apiKey = m.getApiKey();myModel.apiUrl = m.getApiUrl();myModel.type = m.getType();myModel.status = m.getStatus();myModel.params = m.getParams();// 这里只实现Ollama，后续可扩展其他供应商if ("ollama".equalsIgnoreCase(m.getVendor())) {OllamaApi ollamaApi = OllamaApi.builder().baseUrl(m.getApiUrl()).build();// 解析参数OllamaOptions.Builder optionsBuilder = OllamaOptions.builder().model(m.getName());if (m.getParams() != null && !m.getParams().isEmpty()) {try {Map<String, Object> paramMap = objectMapper.readValue(m.getParams(), Map.class);if (paramMap.containsKey("temperature")) {optionsBuilder.temperature(Double.parseDouble(paramMap.get("temperature").toString()));}// 可扩展更多参数} catch (Exception ignored) {}}myModel.chatModel = OllamaChatModel.builder().ollamaApi(ollamaApi).defaultOptions(optionsBuilder.build()).build();}// TODO: 其他供应商实现modelHashMap.put(m.getName(), myModel);}}public MyModel getModelByName(String name) {return modelHashMap.get(name);}public void refreshModel(String modelName) {AiModel m = aiModelRepository.findByName(modelName);if (m == null) return;MyModel myModel = new MyModel();myModel.vendor = m.getVendor();myModel.name = m.getName();myModel.apiKey = m.getApiKey();myModel.apiUrl = m.getApiUrl();myModel.type = m.getType();myModel.status = m.getStatus();myModel.params = m.getParams();if ("ollama".equalsIgnoreCase(m.getVendor())) {OllamaApi ollamaApi = OllamaApi.builder().baseUrl(m.getApiUrl()).build();OllamaOptions.Builder optionsBuilder = OllamaOptions.builder().model(m.getName());if (m.getParams() != null && !m.getParams().isEmpty()) {try {Map<String, Object> paramMap = new ObjectMapper().readValue(m.getParams(), Map.class);if (paramMap.containsKey("temperature")) {optionsBuilder.temperature(Double.parseDouble(paramMap.get("temperature").toString()));}} catch (Exception ignored) {}}myModel.chatModel = OllamaChatModel.builder().ollamaApi(ollamaApi).defaultOptions(optionsBuilder.build()).build();}// TODO: 其他供应商实现modelHashMap.put(m.getName(), myModel);}public static class MyModel {@Schema(description = "供应商")private String vendor;@Schema(description = "模型名称")private String name;@Schema(description = "密钥")private String apiKey;@Schema(description = "模型API地址")private String apiUrl;@Schema(description = "类型（对话、图片、音频、视频、量化）")private String type;@Schema(description = "模型可用状态 1:可用 0:不可用")private Integer status;@Schema(description = "模型参数（如温度等，json格式）")private String params;@Schema(description = "ollama对应的会话对象")private OllamaChatModel chatModel;// TODO: 其他供应商的会话ChatModelpublic OllamaChatModel getChatModel() { return chatModel; }public String getVendor() { return vendor; }public String getName() { return name; }public String getApiKey() { return apiKey; }public String getApiUrl() { return apiUrl; }public String getType() { return type; }public Integer getStatus() { return status; }public String getParams() { return params; }}
}
 ✨ | ThemeClock（主题时钟）)
