seeseemelk/ollama4j
1
0
Fork 0
forked from Mirror/ollama4j
Code Issues Pull Requests Actions Packages Releases 1 Activity

Merge remote-tracking branch 'origin/main'

Browse Source
This commit is contained in:
Amith Koujalgi 2023-12-14 17:30:27 +05:30
commit f6af7025a8
6 changed files with 558 additions and 378 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
README.md
View File

@ -91,7 +91,13 @@ For simplest way to get started, I prefer to use the Ollama docker setup.
Start the Ollama docker container:
```shell
docker run -v ~/ollama:/root/.ollama -p 11434:11434 ollama/ollama
docker run -it -v ~/ollama:/root/.ollama -p 11434:11434 ollama/ollama
```
With GPUs
```shell
docker run -it --gpus=all -v ~/ollama:/root/.ollama -p 11434:11434 ollama/ollama
```
Instantiate `OllamaAPI`
@ -347,6 +353,7 @@ Find the full `Javadoc` (API specifications) [here](https://amithkoujalgi.github
conversational memory
- `stream`: Add support for streaming responses from the model
- [x] Setup logging
- [ ] Use lombok
- [ ] Add test cases
- [ ] Handle exceptions better (maybe throw more appropriate exceptions)

149
src/main/java/io/github/amithkoujalgi/ollama4j/core/OllamaAPI.java
View File

@ -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;
@ -17,10 +14,10 @@ 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 {
@ -43,6 +40,7 @@ public class OllamaAPI {
/**
* Set/unset logging of responses
*
* @param verbose true/false
*/
public void setVerbose(boolean verbose) {
@ -54,38 +52,59 @@ public class OllamaAPI {
*
* @return the list
*/
public List<Model> listModels() throws OllamaBaseException, IOException, InterruptedException, URISyntaxException {
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());
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();
return Utils.getObjectMapper()
.readValue(responseString, ListModelsResponse.class)
.getModels();
} else {
throw new OllamaBaseException(statusCode + " - " + responseString);
}
}
/**
* Pull a model on the Ollama server from the list of <a href="https://ollama.ai/library">available models</a>.
* Pull a model on the Ollama server from the list of <a
* href="https://ollama.ai/library">available models</a>.
*
* @param model the name of the model
*/
public void pullModel(String model) throws OllamaBaseException, IOException, URISyntaxException, InterruptedException {
public void pullModel(String model)
throws OllamaBaseException, IOException, URISyntaxException, InterruptedException {
String url = this.host + "/api/pull";
String jsonData = String.format("{\"name\": \"%s\"}", model);
HttpRequest request = HttpRequest.newBuilder().uri(new URI(url)).POST(HttpRequest.BodyPublishers.ofString(jsonData)).header("Accept", "application/json").header("Content-type", "application/json").build();
HttpRequest request =
HttpRequest.newBuilder()
.uri(new URI(url))
.POST(HttpRequest.BodyPublishers.ofString(jsonData))
.header("Accept", "application/json")
.header("Content-type", "application/json")
.build();
HttpClient client = HttpClient.newHttpClient();
HttpResponse<InputStream> response = client.send(request, HttpResponse.BodyHandlers.ofInputStream());
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))) {
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);
ModelPullResponse modelPullResponse =
Utils.getObjectMapper().readValue(line, ModelPullResponse.class);
if (verbose) {
logger.info(modelPullResponse.getStatus());
}
@ -102,10 +121,17 @@ public class OllamaAPI {
* @param modelName the model
* @return the model details
*/
public ModelDetail getModelDetails(String modelName) throws IOException, OllamaBaseException, InterruptedException {
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();
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();
@ -118,16 +144,24 @@ public class OllamaAPI {
}
/**
* 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>.
* 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 {
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();
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();
@ -135,7 +169,8 @@ public class OllamaAPI {
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.
// 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);
}
@ -148,12 +183,20 @@ public class OllamaAPI {
* 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.
* @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 {
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();
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();
@ -173,16 +216,24 @@ public class OllamaAPI {
* @param prompt text to generate embeddings for
* @return embeddings
*/
public List<Double> generateEmbeddings(String model, String prompt) throws IOException, InterruptedException, OllamaBaseException {
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();
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);
EmbeddingResponse embeddingResponse =
Utils.getObjectMapper().readValue(responseBody, EmbeddingResponse.class);
return embeddingResponse.getEmbedding();
} else {
throw new OllamaBaseException(statusCode + " - " + responseBody);
@ -196,36 +247,53 @@ public class OllamaAPI {
* @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 {
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());
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))) {
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 (statusCode == 404) {
OllamaErrorResponseModel ollamaResponseModel =
Utils.getObjectMapper().readValue(line, OllamaErrorResponseModel.class);
responseBuffer.append(ollamaResponseModel.getError());
} else {
OllamaResponseModel ollamaResponseModel =
Utils.getObjectMapper().readValue(line, OllamaResponseModel.class);
if (!ollamaResponseModel.getDone()) {
responseBuffer.append(ollamaResponseModel.getResponse());
}
}
}
}
if (statusCode != 200) {
throw new OllamaBaseException(statusCode + " - " + responseBuffer);
throw new OllamaBaseException(responseBuffer.toString());
} else {
long endTime = System.currentTimeMillis();
return new OllamaResult(responseBuffer.toString().trim(), endTime - startTime);
return new OllamaResult(responseBuffer.toString().trim(), endTime - startTime, statusCode);
}
}
/**
* 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.
* 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
@ -235,7 +303,8 @@ public class OllamaAPI {
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 ollamaAsyncResultCallback =
new OllamaAsyncResultCallback(httpClient, uri, ollamaRequestModel);
ollamaAsyncResultCallback.start();
return ollamaAsyncResultCallback;
}

73
src/main/java/io/github/amithkoujalgi/ollama4j/core/models/OllamaAsyncResultCallback.java
View File

@ -2,7 +2,6 @@ package io.github.amithkoujalgi.ollama4j.core.models;
import io.github.amithkoujalgi.ollama4j.core.exceptions.OllamaBaseException;
import io.github.amithkoujalgi.ollama4j.core.utils.Utils;
import java.io.BufferedReader;
import java.io.IOException;
import java.io.InputStream;
@ -23,9 +22,13 @@ public class OllamaAsyncResultCallback extends Thread {
private final Queue<String> queue = new LinkedList<>();
private String result;
private boolean isDone;
private boolean succeeded;
private int httpStatusCode;
private long responseTime = 0;
public OllamaAsyncResultCallback(HttpClient client, URI uri, OllamaRequestModel ollamaRequestModel) {
public OllamaAsyncResultCallback(
HttpClient client, URI uri, OllamaRequestModel ollamaRequestModel) {
this.client = client;
this.ollamaRequestModel = ollamaRequestModel;
this.uri = uri;
@ -38,44 +41,89 @@ public class OllamaAsyncResultCallback extends Thread {
public void run() {
try {
long startTime = System.currentTimeMillis();
HttpRequest request = HttpRequest.newBuilder(uri).POST(HttpRequest.BodyPublishers.ofString(Utils.getObjectMapper().writeValueAsString(ollamaRequestModel))).header("Content-Type", "application/json").build();
HttpResponse<InputStream> response = client.send(request, HttpResponse.BodyHandlers.ofInputStream());
HttpRequest request =
HttpRequest.newBuilder(uri)
.POST(
HttpRequest.BodyPublishers.ofString(
Utils.getObjectMapper().writeValueAsString(ollamaRequestModel)))
.header("Content-Type", "application/json")
.build();
HttpResponse<InputStream> response =
client.send(request, HttpResponse.BodyHandlers.ofInputStream());
int statusCode = response.statusCode();
this.httpStatusCode = statusCode;
InputStream responseBodyStream = response.body();
String responseString = "";
try (BufferedReader reader = new BufferedReader(new InputStreamReader(responseBodyStream, StandardCharsets.UTF_8))) {
try (BufferedReader reader =
new BufferedReader(new InputStreamReader(responseBodyStream, StandardCharsets.UTF_8))) {
String line;
StringBuilder responseBuffer = new StringBuilder();
while ((line = reader.readLine()) != null) {
OllamaResponseModel ollamaResponseModel = Utils.getObjectMapper().readValue(line, OllamaResponseModel.class);
if (statusCode == 404) {
OllamaErrorResponseModel ollamaResponseModel =
Utils.getObjectMapper().readValue(line, OllamaErrorResponseModel.class);
queue.add(ollamaResponseModel.getError());
responseBuffer.append(ollamaResponseModel.getError());
} else {
OllamaResponseModel ollamaResponseModel =
Utils.getObjectMapper().readValue(line, OllamaResponseModel.class);
queue.add(ollamaResponseModel.getResponse());
if (!ollamaResponseModel.getDone()) {
responseBuffer.append(ollamaResponseModel.getResponse());
}
}
}
reader.close();
this.isDone = true;
this.succeeded = true;
this.result = responseBuffer.toString();
long endTime = System.currentTimeMillis();
responseTime = endTime - startTime;
}
if (statusCode != 200) {
throw new OllamaBaseException(statusCode + " - " + responseString);
throw new OllamaBaseException(this.result);
}
} catch (IOException | InterruptedException | OllamaBaseException e) {
this.isDone = true;
this.result = "FAILED! " + e.getMessage();
this.succeeded = false;
this.result = "[FAILED] " + e.getMessage();
}
}
/**
* Returns the status of the thread. This does not indicate that the request was successful or a
* failure, rather it is just a status flag to indicate if the thread is active or ended.
*
* @return boolean - status
*/
public boolean isComplete() {
return isDone;
}
/**
* Returns the HTTP response status code for the request that was made to Ollama server.
*
* @return int - the status code for the request
*/
public int getHttpStatusCode() {
return httpStatusCode;
}
/**
* Returns the status of the request. Indicates if the request was successful or a failure. If the
* request was a failure, the `getResponse()` method will return the error message.
*
* @return boolean - status
*/
public boolean isSucceeded() {
return succeeded;
}
/**
* Returns the final response when the execution completes. Does not return intermediate results.
* @return response text
*
* @return String - response text
*/
public String getResponse() {
return result;
@ -86,8 +134,9 @@ public class OllamaAsyncResultCallback extends Thread {
}
/**
* Returns the response time in seconds.
* @return response time in seconds
* Returns the response time in milliseconds.
*
* @return long - response time in milliseconds.
*/
public long getResponseTime() {
return responseTime;

18
src/main/java/io/github/amithkoujalgi/ollama4j/core/models/OllamaErrorResponseModel.java Normal file
View File

@ -0,0 +1,18 @@
package io.github.amithkoujalgi.ollama4j.core.models;
import com.fasterxml.jackson.annotation.JsonIgnoreProperties;
import com.fasterxml.jackson.annotation.JsonProperty;
import java.util.List;
@JsonIgnoreProperties(ignoreUnknown = true)
public class OllamaErrorResponseModel {
private String error;
public String getError() {
return error;
}
public void setError(String error) {
this.error = error;
}
}

41
src/main/java/io/github/amithkoujalgi/ollama4j/core/models/OllamaResult.java
View File

@ -1,20 +1,57 @@
package io.github.amithkoujalgi.ollama4j.core.models;
import static io.github.amithkoujalgi.ollama4j.core.utils.Utils.getObjectMapper;
import com.fasterxml.jackson.core.JsonProcessingException;
/** The type Ollama result. */
@SuppressWarnings("unused")
public class OllamaResult {
private String response;
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;
}
/**
* Get the response text
*
* @return String - response text
*/
public String getResponse() {
return response;
}
/**
* Get the response time in milliseconds.
*
* @return long - response time in milliseconds
*/
public long getResponseTime() {
return responseTime;
}
/**
* Get the response status code.
*
* @return int - response status code
*/
public int getHttpStatusCode() {
return httpStatusCode;
}
@Override
public String toString() {
try {
return getObjectMapper().writerWithDefaultPrettyPrinter().writeValueAsString(this);
} catch (JsonProcessingException e) {
throw new RuntimeException(e);
}
}
}

14
src/test/java/io/github/amithkoujalgi/ollama4j/unittests/TestMockedAPIs.java
View File

@ -1,19 +1,18 @@
package io.github.amithkoujalgi.ollama4j.unittests;
import static org.mockito.Mockito.*;
import io.github.amithkoujalgi.ollama4j.core.OllamaAPI;
import io.github.amithkoujalgi.ollama4j.core.exceptions.OllamaBaseException;
import io.github.amithkoujalgi.ollama4j.core.models.ModelDetail;
import io.github.amithkoujalgi.ollama4j.core.models.OllamaAsyncResultCallback;
import io.github.amithkoujalgi.ollama4j.core.models.OllamaResult;
import io.github.amithkoujalgi.ollama4j.core.types.OllamaModelType;
import org.junit.jupiter.api.Test;
import org.mockito.Mockito;
import java.io.IOException;
import java.net.URISyntaxException;
import java.util.ArrayList;
import static org.mockito.Mockito.*;
import org.junit.jupiter.api.Test;
import org.mockito.Mockito;
public class TestMockedAPIs {
@Test
@ -101,7 +100,7 @@ public class TestMockedAPIs {
String model = OllamaModelType.LLAMA2;
String prompt = "some prompt text";
try {
when(ollamaAPI.ask(model, prompt)).thenReturn(new OllamaResult("", 0));
when(ollamaAPI.ask(model, prompt)).thenReturn(new OllamaResult("", 0, 200));
ollamaAPI.ask(model, prompt);
verify(ollamaAPI, times(1)).ask(model, prompt);
} catch (IOException | OllamaBaseException | InterruptedException e) {
@ -114,7 +113,8 @@ public class TestMockedAPIs {
OllamaAPI ollamaAPI = Mockito.mock(OllamaAPI.class);
String model = OllamaModelType.LLAMA2;
String prompt = "some prompt text";
when(ollamaAPI.askAsync(model, prompt)).thenReturn(new OllamaAsyncResultCallback(null, null, null));
when(ollamaAPI.askAsync(model, prompt))
.thenReturn(new OllamaAsyncResultCallback(null, null, null));
ollamaAPI.askAsync(model, prompt);
verify(ollamaAPI, times(1)).askAsync(model, prompt);
}