Fixes to ask API

src/main/java/io/github/amithkoujalgi/ollama4j/core/OllamaAPI.java

@@ -3,9 +3,6 @@ package io.github.amithkoujalgi.ollama4j.core;
 import io.github.amithkoujalgi.ollama4j.core.exceptions.OllamaBaseException;
 import io.github.amithkoujalgi.ollama4j.core.models.*;
 import io.github.amithkoujalgi.ollama4j.core.utils.Utils;
-import org.slf4j.Logger;
-import org.slf4j.LoggerFactory;
-
 import java.io.BufferedReader;
 import java.io.IOException;
 import java.io.InputStream;
@@ -17,226 +14,298 @@ import java.net.http.HttpRequest;
 import java.net.http.HttpResponse;
 import java.nio.charset.StandardCharsets;
 import java.util.List;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
 
-/**
+/** The base Ollama API class. */
- * The base Ollama API class.
- */
 @SuppressWarnings("DuplicatedCode")
 public class OllamaAPI {
 
-    private static final Logger logger = LoggerFactory.getLogger(OllamaAPI.class);
+  private static final Logger logger = LoggerFactory.getLogger(OllamaAPI.class);
-    private final String host;
+  private final String host;
-    private boolean verbose = false;
+  private boolean verbose = false;
 
-    /**
+  /**
-     * Instantiates the Ollama API.
+   * Instantiates the Ollama API.
-     *
+   *
-     * @param host the host address of Ollama server
+   * @param host the host address of Ollama server
-     */
+   */
-    public OllamaAPI(String host) {
+  public OllamaAPI(String host) {
-        if (host.endsWith("/")) {
+    if (host.endsWith("/")) {
-            this.host = host.substring(0, host.length() - 1);
+      this.host = host.substring(0, host.length() - 1);
-        } else {
+    } else {
-            this.host = host;
+      this.host = host;
-        }
     }
+  }
 
-    /**
+  /**
-     * Set/unset logging of responses
+   * Set/unset logging of responses
-     * @param verbose true/false
+   *
-     */
+   * @param verbose true/false
-    public void setVerbose(boolean verbose) {
+   */
-        this.verbose = verbose;
+  public void setVerbose(boolean verbose) {
+    this.verbose = verbose;
+  }
+
+  /**
+   * List available models from Ollama server.
+   *
+   * @return the list
+   */
+  public List<Model> listModels()
+      throws OllamaBaseException, IOException, InterruptedException, URISyntaxException {
+    String url = this.host + "/api/tags";
+    HttpClient httpClient = HttpClient.newHttpClient();
+    HttpRequest httpRequest =
+        HttpRequest.newBuilder()
+            .uri(new URI(url))
+            .header("Accept", "application/json")
+            .header("Content-type", "application/json")
+            .GET()
+            .build();
+    HttpResponse<String> response =
+        httpClient.send(httpRequest, HttpResponse.BodyHandlers.ofString());
+    int statusCode = response.statusCode();
+    String responseString = response.body();
+    if (statusCode == 200) {
+      return Utils.getObjectMapper()
+          .readValue(responseString, ListModelsResponse.class)
+          .getModels();
+    } else {
+      throw new OllamaBaseException(statusCode + " - " + responseString);
     }
+  }
 
-    /**
+  /**
-     * List available models from Ollama server.
+   * Pull a model on the Ollama server from the list of <a
-     *
+   * href="https://ollama.ai/library">available models</a>.
-     * @return the list
+   *
-     */
+   * @param model the name of the model
-    public List<Model> listModels() throws OllamaBaseException, IOException, InterruptedException, URISyntaxException {
+   */
-        String url = this.host + "/api/tags";
+  public void pullModel(String model)
-        HttpClient httpClient = HttpClient.newHttpClient();
+      throws OllamaBaseException, IOException, URISyntaxException, InterruptedException {
-        HttpRequest httpRequest = HttpRequest.newBuilder().uri(new URI(url)).header("Accept", "application/json").header("Content-type", "application/json").GET().build();
+    String url = this.host + "/api/pull";
-        HttpResponse<String> response = httpClient.send(httpRequest, HttpResponse.BodyHandlers.ofString());
+    String jsonData = String.format("{\"name\": \"%s\"}", model);
-        int statusCode = response.statusCode();
+    HttpRequest request =
-        String responseString = response.body();
+        HttpRequest.newBuilder()
-        if (statusCode == 200) {
+            .uri(new URI(url))
-            return Utils.getObjectMapper().readValue(responseString, ListModelsResponse.class).getModels();
+            .POST(HttpRequest.BodyPublishers.ofString(jsonData))
-        } else {
+            .header("Accept", "application/json")
-            throw new OllamaBaseException(statusCode + " - " + responseString);
+            .header("Content-type", "application/json")
-        }
+            .build();
-    }
+    HttpClient client = HttpClient.newHttpClient();
-
+    HttpResponse<InputStream> response =
-    /**
+        client.send(request, HttpResponse.BodyHandlers.ofInputStream());
-     * Pull a model on the Ollama server from the list of <a href="https://ollama.ai/library">available models</a>.
+    int statusCode = response.statusCode();
-     *
+    InputStream responseBodyStream = response.body();
-     * @param model the name of the model
+    String responseString = "";
-     */
+    try (BufferedReader reader =
-    public void pullModel(String model) throws OllamaBaseException, IOException, URISyntaxException, InterruptedException {
+        new BufferedReader(new InputStreamReader(responseBodyStream, StandardCharsets.UTF_8))) {
-        String url = this.host + "/api/pull";
+      String line;
-        String jsonData = String.format("{\"name\": \"%s\"}", model);
+      while ((line = reader.readLine()) != null) {
-        HttpRequest request = HttpRequest.newBuilder().uri(new URI(url)).POST(HttpRequest.BodyPublishers.ofString(jsonData)).header("Accept", "application/json").header("Content-type", "application/json").build();
+        ModelPullResponse modelPullResponse =
-        HttpClient client = HttpClient.newHttpClient();
+            Utils.getObjectMapper().readValue(line, ModelPullResponse.class);
-        HttpResponse<InputStream> response = client.send(request, HttpResponse.BodyHandlers.ofInputStream());
-        int statusCode = response.statusCode();
-        InputStream responseBodyStream = response.body();
-        String responseString = "";
-        try (BufferedReader reader = new BufferedReader(new InputStreamReader(responseBodyStream, StandardCharsets.UTF_8))) {
-            String line;
-            while ((line = reader.readLine()) != null) {
-                ModelPullResponse modelPullResponse = Utils.getObjectMapper().readValue(line, ModelPullResponse.class);
-                if (verbose) {
-                    logger.info(modelPullResponse.getStatus());
-                }
-            }
-        }
-        if (statusCode != 200) {
-            throw new OllamaBaseException(statusCode + " - " + responseString);
-        }
-    }
-
-    /**
-     * Gets model details from the Ollama server.
-     *
-     * @param modelName the model
-     * @return the model details
-     */
-    public ModelDetail getModelDetails(String modelName) throws IOException, OllamaBaseException, InterruptedException {
-        String url = this.host + "/api/show";
-        String jsonData = String.format("{\"name\": \"%s\"}", modelName);
-        HttpRequest request = HttpRequest.newBuilder().uri(URI.create(url)).header("Accept", "application/json").header("Content-type", "application/json").POST(HttpRequest.BodyPublishers.ofString(jsonData)).build();
-        HttpClient client = HttpClient.newHttpClient();
-        HttpResponse<String> response = client.send(request, HttpResponse.BodyHandlers.ofString());
-        int statusCode = response.statusCode();
-        String responseBody = response.body();
-        if (statusCode == 200) {
-            return Utils.getObjectMapper().readValue(responseBody, ModelDetail.class);
-        } else {
-            throw new OllamaBaseException(statusCode + " - " + responseBody);
-        }
-    }
-
-    /**
-     * Create a custom model from a model file.
-     * Read more about custom model file creation <a href="https://github.com/jmorganca/ollama/blob/main/docs/modelfile.md">here</a>.
-     *
-     * @param modelName the name of the custom model to be created.
-     * @param modelFilePath the path to model file that exists on the Ollama server.
-     */
-    public void createModel(String modelName, String modelFilePath) throws IOException, InterruptedException, OllamaBaseException {
-        String url = this.host + "/api/create";
-        String jsonData = String.format("{\"name\": \"%s\", \"path\": \"%s\"}", modelName, modelFilePath);
-        HttpRequest request = HttpRequest.newBuilder().uri(URI.create(url)).header("Accept", "application/json").header("Content-Type", "application/json").POST(HttpRequest.BodyPublishers.ofString(jsonData, StandardCharsets.UTF_8)).build();
-        HttpClient client = HttpClient.newHttpClient();
-        HttpResponse<String> response = client.send(request, HttpResponse.BodyHandlers.ofString());
-        int statusCode = response.statusCode();
-        String responseString = response.body();
-        if (statusCode != 200) {
-            throw new OllamaBaseException(statusCode + " - " + responseString);
-        }
-        // FIXME: Ollama API returns HTTP status code 200 for model creation failure cases. Correct this if the issue is fixed in the Ollama API server.
-        if (responseString.contains("error")) {
-            throw new OllamaBaseException(responseString);
-        }
         if (verbose) {
-            logger.info(responseString);
+          logger.info(modelPullResponse.getStatus());
         }
+      }
     }
-
+    if (statusCode != 200) {
-    /**
+      throw new OllamaBaseException(statusCode + " - " + responseString);
-     * Delete a model from Ollama server.
-     *
-     * @param name the name of the model to be deleted.
-     * @param ignoreIfNotPresent - ignore errors if the specified model is not present on Ollama server.
-     */
-    public void deleteModel(String name, boolean ignoreIfNotPresent) throws IOException, InterruptedException, OllamaBaseException {
-        String url = this.host + "/api/delete";
-        String jsonData = String.format("{\"name\": \"%s\"}", name);
-        HttpRequest request = HttpRequest.newBuilder().uri(URI.create(url)).method("DELETE", HttpRequest.BodyPublishers.ofString(jsonData, StandardCharsets.UTF_8)).header("Accept", "application/json").header("Content-type", "application/json").build();
-        HttpClient client = HttpClient.newHttpClient();
-        HttpResponse<String> response = client.send(request, HttpResponse.BodyHandlers.ofString());
-        int statusCode = response.statusCode();
-        String responseBody = response.body();
-        if (statusCode == 404 && responseBody.contains("model") && responseBody.contains("not found")) {
-            return;
-        }
-        if (statusCode != 200) {
-            throw new OllamaBaseException(statusCode + " - " + responseBody);
-        }
     }
+  }
 
-    /**
+  /**
-     * Generate embeddings for a given text from a model
+   * Gets model details from the Ollama server.
-     *
+   *
-     * @param model name of model to generate embeddings from
+   * @param modelName the model
-     * @param prompt text to generate embeddings for
+   * @return the model details
-     * @return embeddings
+   */
-     */
+  public ModelDetail getModelDetails(String modelName)
-    public List<Double> generateEmbeddings(String model, String prompt) throws IOException, InterruptedException, OllamaBaseException {
+      throws IOException, OllamaBaseException, InterruptedException {
-        String url = this.host + "/api/embeddings";
+    String url = this.host + "/api/show";
-        String jsonData = String.format("{\"model\": \"%s\", \"prompt\": \"%s\"}", model, prompt);
+    String jsonData = String.format("{\"name\": \"%s\"}", modelName);
-        HttpClient httpClient = HttpClient.newHttpClient();
+    HttpRequest request =
-        HttpRequest request = HttpRequest.newBuilder().uri(URI.create(url)).header("Accept", "application/json").header("Content-type", "application/json").POST(HttpRequest.BodyPublishers.ofString(jsonData)).build();
+        HttpRequest.newBuilder()
-        HttpResponse<String> response = httpClient.send(request, HttpResponse.BodyHandlers.ofString());
+            .uri(URI.create(url))
-        int statusCode = response.statusCode();
+            .header("Accept", "application/json")
-        String responseBody = response.body();
+            .header("Content-type", "application/json")
-        if (statusCode == 200) {
+            .POST(HttpRequest.BodyPublishers.ofString(jsonData))
-            EmbeddingResponse embeddingResponse = Utils.getObjectMapper().readValue(responseBody, EmbeddingResponse.class);
+            .build();
-            return embeddingResponse.getEmbedding();
+    HttpClient client = HttpClient.newHttpClient();
+    HttpResponse<String> response = client.send(request, HttpResponse.BodyHandlers.ofString());
+    int statusCode = response.statusCode();
+    String responseBody = response.body();
+    if (statusCode == 200) {
+      return Utils.getObjectMapper().readValue(responseBody, ModelDetail.class);
+    } else {
+      throw new OllamaBaseException(statusCode + " - " + responseBody);
+    }
+  }
+
+  /**
+   * Create a custom model from a model file. Read more about custom model file creation <a
+   * href="https://github.com/jmorganca/ollama/blob/main/docs/modelfile.md">here</a>.
+   *
+   * @param modelName the name of the custom model to be created.
+   * @param modelFilePath the path to model file that exists on the Ollama server.
+   */
+  public void createModel(String modelName, String modelFilePath)
+      throws IOException, InterruptedException, OllamaBaseException {
+    String url = this.host + "/api/create";
+    String jsonData =
+        String.format("{\"name\": \"%s\", \"path\": \"%s\"}", modelName, modelFilePath);
+    HttpRequest request =
+        HttpRequest.newBuilder()
+            .uri(URI.create(url))
+            .header("Accept", "application/json")
+            .header("Content-Type", "application/json")
+            .POST(HttpRequest.BodyPublishers.ofString(jsonData, StandardCharsets.UTF_8))
+            .build();
+    HttpClient client = HttpClient.newHttpClient();
+    HttpResponse<String> response = client.send(request, HttpResponse.BodyHandlers.ofString());
+    int statusCode = response.statusCode();
+    String responseString = response.body();
+    if (statusCode != 200) {
+      throw new OllamaBaseException(statusCode + " - " + responseString);
+    }
+    // FIXME: Ollama API returns HTTP status code 200 for model creation failure cases. Correct this
+    // if the issue is fixed in the Ollama API server.
+    if (responseString.contains("error")) {
+      throw new OllamaBaseException(responseString);
+    }
+    if (verbose) {
+      logger.info(responseString);
+    }
+  }
+
+  /**
+   * Delete a model from Ollama server.
+   *
+   * @param name the name of the model to be deleted.
+   * @param ignoreIfNotPresent - ignore errors if the specified model is not present on Ollama
+   *     server.
+   */
+  public void deleteModel(String name, boolean ignoreIfNotPresent)
+      throws IOException, InterruptedException, OllamaBaseException {
+    String url = this.host + "/api/delete";
+    String jsonData = String.format("{\"name\": \"%s\"}", name);
+    HttpRequest request =
+        HttpRequest.newBuilder()
+            .uri(URI.create(url))
+            .method("DELETE", HttpRequest.BodyPublishers.ofString(jsonData, StandardCharsets.UTF_8))
+            .header("Accept", "application/json")
+            .header("Content-type", "application/json")
+            .build();
+    HttpClient client = HttpClient.newHttpClient();
+    HttpResponse<String> response = client.send(request, HttpResponse.BodyHandlers.ofString());
+    int statusCode = response.statusCode();
+    String responseBody = response.body();
+    if (statusCode == 404 && responseBody.contains("model") && responseBody.contains("not found")) {
+      return;
+    }
+    if (statusCode != 200) {
+      throw new OllamaBaseException(statusCode + " - " + responseBody);
+    }
+  }
+
+  /**
+   * Generate embeddings for a given text from a model
+   *
+   * @param model name of model to generate embeddings from
+   * @param prompt text to generate embeddings for
+   * @return embeddings
+   */
+  public List<Double> generateEmbeddings(String model, String prompt)
+      throws IOException, InterruptedException, OllamaBaseException {
+    String url = this.host + "/api/embeddings";
+    String jsonData = String.format("{\"model\": \"%s\", \"prompt\": \"%s\"}", model, prompt);
+    HttpClient httpClient = HttpClient.newHttpClient();
+    HttpRequest request =
+        HttpRequest.newBuilder()
+            .uri(URI.create(url))
+            .header("Accept", "application/json")
+            .header("Content-type", "application/json")
+            .POST(HttpRequest.BodyPublishers.ofString(jsonData))
+            .build();
+    HttpResponse<String> response = httpClient.send(request, HttpResponse.BodyHandlers.ofString());
+    int statusCode = response.statusCode();
+    String responseBody = response.body();
+    if (statusCode == 200) {
+      EmbeddingResponse embeddingResponse =
+          Utils.getObjectMapper().readValue(responseBody, EmbeddingResponse.class);
+      return embeddingResponse.getEmbedding();
+    } else {
+      throw new OllamaBaseException(statusCode + " - " + responseBody);
+    }
+  }
+
+  /**
+   * Ask a question to a model running on Ollama server. This is a sync/blocking call.
+   *
+   * @param ollamaModelType the ollama model to ask the question to
+   * @param promptText the prompt/question text
+   * @return OllamaResult - that includes response text and time taken for response
+   */
+  public OllamaResult ask(String ollamaModelType, String promptText)
+      throws OllamaBaseException, IOException, InterruptedException {
+    OllamaRequestModel ollamaRequestModel = new OllamaRequestModel(ollamaModelType, promptText);
+    long startTime = System.currentTimeMillis();
+    HttpClient httpClient = HttpClient.newHttpClient();
+    URI uri = URI.create(this.host + "/api/generate");
+    HttpRequest request =
+        HttpRequest.newBuilder(uri)
+            .POST(
+                HttpRequest.BodyPublishers.ofString(
+                    Utils.getObjectMapper().writeValueAsString(ollamaRequestModel)))
+            .header("Content-Type", "application/json")
+            .build();
+    HttpResponse<InputStream> response =
+        httpClient.send(request, HttpResponse.BodyHandlers.ofInputStream());
+    int statusCode = response.statusCode();
+    InputStream responseBodyStream = response.body();
+    StringBuilder responseBuffer = new StringBuilder();
+    try (BufferedReader reader =
+        new BufferedReader(new InputStreamReader(responseBodyStream, StandardCharsets.UTF_8))) {
+      String line;
+      while ((line = reader.readLine()) != null) {
+        if (statusCode == 404) {
+          OllamaErrorResponseModel ollamaResponseModel =
+              Utils.getObjectMapper().readValue(line, OllamaErrorResponseModel.class);
+          responseBuffer.append(ollamaResponseModel.getError());
         } else {
-            throw new OllamaBaseException(statusCode + " - " + responseBody);
+          OllamaResponseModel ollamaResponseModel =
+              Utils.getObjectMapper().readValue(line, OllamaResponseModel.class);
+          if (!ollamaResponseModel.getDone()) {
+            responseBuffer.append(ollamaResponseModel.getResponse());
+          }
         }
+      }
     }
+    if (statusCode != 200) {
+      throw new OllamaBaseException(responseBuffer.toString());
+    } else {
+      long endTime = System.currentTimeMillis();
+      return new OllamaResult(responseBuffer.toString().trim(), endTime - startTime, statusCode);
+    }
+  }
 
-    /**
+  /**
-     * Ask a question to a model running on Ollama server. This is a sync/blocking call.
+   * Ask a question to a model running on Ollama server and get a callback handle that can be used
-     *
+   * to check for status and get the response from the model later. This would be an
-     * @param ollamaModelType the ollama model to ask the question to
+   * async/non-blocking call.
-     * @param promptText the prompt/question text
+   *
-     * @return OllamaResult - that includes response text and time taken for response
+   * @param ollamaModelType the ollama model to ask the question to
-     */
+   * @param promptText the prompt/question text
-    public OllamaResult ask(String ollamaModelType, String promptText) throws OllamaBaseException, IOException, InterruptedException {
+   * @return the ollama async result callback handle
-        OllamaRequestModel ollamaRequestModel = new OllamaRequestModel(ollamaModelType, promptText);
+   */
-        long startTime = System.currentTimeMillis();
+  public OllamaAsyncResultCallback askAsync(String ollamaModelType, String promptText) {
-        HttpClient httpClient = HttpClient.newHttpClient();
+    OllamaRequestModel ollamaRequestModel = new OllamaRequestModel(ollamaModelType, promptText);
-        URI uri = URI.create(this.host + "/api/generate");
+    HttpClient httpClient = HttpClient.newHttpClient();
-        HttpRequest request = HttpRequest.newBuilder(uri).POST(HttpRequest.BodyPublishers.ofString(Utils.getObjectMapper().writeValueAsString(ollamaRequestModel))).header("Content-Type", "application/json").build();
+    URI uri = URI.create(this.host + "/api/generate");
-        HttpResponse<InputStream> response = httpClient.send(request, HttpResponse.BodyHandlers.ofInputStream());
+    OllamaAsyncResultCallback ollamaAsyncResultCallback =
-        int statusCode = response.statusCode();
+        new OllamaAsyncResultCallback(httpClient, uri, ollamaRequestModel);
-        InputStream responseBodyStream = response.body();
+    ollamaAsyncResultCallback.start();
-        StringBuilder responseBuffer = new StringBuilder();
+    return ollamaAsyncResultCallback;
-        try (BufferedReader reader = new BufferedReader(new InputStreamReader(responseBodyStream, StandardCharsets.UTF_8))) {
+  }
-            String line;
-            while ((line = reader.readLine()) != null) {
-                OllamaResponseModel ollamaResponseModel = Utils.getObjectMapper().readValue(line, OllamaResponseModel.class);
-                if (!ollamaResponseModel.getDone()) {
-                    responseBuffer.append(ollamaResponseModel.getResponse());
-                }
-            }
-        }
-        if (statusCode != 200) {
-            throw new OllamaBaseException(statusCode + " - " + responseBuffer);
-        } else {
-            long endTime = System.currentTimeMillis();
-            return new OllamaResult(responseBuffer.toString().trim(), endTime - startTime);
-        }
-    }
-
-    /**
-     * Ask a question to a model running on Ollama server and get a callback handle that can be used to check for status and get the response from the model later.
-     * This would be an async/non-blocking call.
-     *
-     * @param ollamaModelType the ollama model to ask the question to
-     * @param promptText the prompt/question text
-     * @return the ollama async result callback handle
-     */
-    public OllamaAsyncResultCallback askAsync(String ollamaModelType, String promptText) {
-        OllamaRequestModel ollamaRequestModel = new OllamaRequestModel(ollamaModelType, promptText);
-        HttpClient httpClient = HttpClient.newHttpClient();
-        URI uri = URI.create(this.host + "/api/generate");
-        OllamaAsyncResultCallback ollamaAsyncResultCallback = new OllamaAsyncResultCallback(httpClient, uri, ollamaRequestModel);
-        ollamaAsyncResultCallback.start();
-        return ollamaAsyncResultCallback;
-    }
 }

src/main/java/io/github/amithkoujalgi/ollama4j/core/models/OllamaResult.java

@@ -9,11 +9,14 @@ import com.fasterxml.jackson.core.JsonProcessingException;
 public class OllamaResult {
   private final String response;
 
+  private int httpStatusCode;
+
   private long responseTime = 0;
 
-  public OllamaResult(String response, long responseTime) {
+  public OllamaResult(String response, long responseTime, int httpStatusCode) {
     this.response = response;
     this.responseTime = responseTime;
+    this.httpStatusCode = httpStatusCode;
   }
 
   /**
@@ -34,6 +37,15 @@ public class OllamaResult {
     return responseTime;
   }
 
+  /**
+   * Get the response status code.
+   *
+   * @return int - response status code
+   */
+  public int getHttpStatusCode() {
+    return httpStatusCode;
+  }
+
   @Override
   public String toString() {
     try {