com.composum.ai.backend.base.service.chat.GPTChatCompletionService Maven / Gradle / Ivy
package com.composum.ai.backend.base.service.chat;
import java.util.List;
import javax.annotation.Nonnull;
import javax.annotation.Nullable;
import com.composum.ai.backend.base.service.GPTException;
/**
* Raw abstraction of the ChatGPT chat interface, with only the details that are needed.
*
* This does deliberately not use the OpenAI API classes because we want to be able to switch to a different API implementation, and hide their complexity from the rest of the code. If we need special parameters,
* we will add new methods with more specific function.
*/
public interface GPTChatCompletionService {
/**
* The simplest case: give some messages and get a single response.
* If the response can be more than a few words, do consider using {@link #streamingChatCompletion(GPTChatRequest, GPTCompletionCallback)} instead,
* to give the user some feedback while waiting.
*/
@Nullable
String getSingleChatCompletion(@Nonnull GPTChatRequest request) throws GPTException;
/**
* Give some messages and receive the streaming response via callback, to reduce waiting time.
* It possibly waits if a rate limit is reached, but otherwise returns immediately after scheduling an asynchronous call.
*/
void streamingChatCompletion(@Nonnull GPTChatRequest request, @Nonnull GPTCompletionCallback callback) throws GPTException;
/**
* Give some messages and receive the streaming response via callback, to reduce waiting time.
* This implementation also performs tool calls if tools are given in {@link GPTChatRequest#getConfiguration()}.
* It possibly waits if a rate limit is reached, but otherwise returns immediately after scheduling an asynchronous call.
*/
void streamingChatCompletionWithToolCalls(@Nonnull GPTChatRequest request, @Nonnull GPTCompletionCallback callback) throws GPTException;
/**
* Retrieves a (usually cached) chat template with that name. Mostly for backend internal use.
* The templates are retrieved from the bundle resources at "chattemplates/", and are cached.
*
* @param templateName the name of the template to retrieve, e.g. "singleTranslation" .
*/
@Nonnull
GPTChatMessagesTemplate getTemplate(@Nonnull String templateName) throws GPTException;
/**
* Helper method to shorten texts by taking out the middle if too long.
* In texts longer than this many tokens we replace the middle with " ... (truncated) ... " since ChatGPT can only
* process a limited number of words / tokens and in the introduction or summary there is probably the most
* condensed information about the text. The output has then maxTokens tokens, including the ... marker.
*
* @param text the text to shorten
* @param maxTokens the maximum number of tokens in the output
*/
@Nonnull
String shorten(@Nullable String text, int maxTokens) throws GPTException;
/**
* Helper for preprocessing HTML so that it can easily read by ChatGPT.
*/
@Nonnull
String htmlToMarkdown(@Nullable String html);
/**
* Opposite of {@link #htmlToMarkdown(String)}.
*/
String markdownToHtml(String markdown);
/**
* Counts the number of tokens for the text for the normally used model. Caution: message boundaries need some tokens
* and slicing text might create a token or two, too, so do not exactly rely on that.
*/
int countTokens(@Nullable String text);
/**
* Whether ChatGPT completion is enabled. If not, calling the methods that access ChatGPT throws an IllegalStateException.
*/
boolean isEnabled();
/**
* Checks whether {@link #isEnabled()} and whether gptConfig enables executing GPT calls.
* (That is currently whether there is an api key either globally or in the gptConfig).
*/
boolean isEnabled(GPTConfiguration gptConfig);
/**
* Returns true if vision is enabled.
*/
boolean isVisionEnabled();
/** Calculates embeddings for the given list of texts. */
@Nonnull
List getEmbeddings(List texts, GPTConfiguration configuration) throws GPTException;
/** Returns the model used for {@link #getEmbeddings(List, GPTConfiguration)}. */
String getEmbeddingsModel();
}