Package com.composum.ai.backend.base.service.chat.impl

Class GPTTranslationServiceImpl

java.lang.Object

com.composum.ai.backend.base.service.chat.impl.GPTTranslationServiceImpl

All Implemented Interfaces:

GPTTranslationService

public class GPTTranslationServiceImpl
extends Object
implements GPTTranslationService

Building on GPTChatCompletionService this implements translation.

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
static interface	GPTTranslationServiceImpl.Config

Field Summary

Fields

Modifier and Type	Field	Description
protected Path	cacheDir
protected GPTChatCompletionService	chatCompletionService
protected GPTTranslationServiceImpl.Config	config
static Pattern	HTML_TAG_AT_START
static String	LASTID
static String	MULTITRANSLATION_SEPARATOR_END	End of separator like `573472 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%` .
protected static Pattern	MULTITRANSLATION_SEPARATOR_PATTERN	Regexp matching separator like `%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 573472 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%` (group "id" matches the number).
static String	MULTITRANSLATION_SEPARATOR_START	Start of separator like `%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 573472 %%%%%%%%%%%%%%%%` .
protected static Pattern	PATTERN_HAS_LETTER
protected static Pattern	PATTERN_SEPARATE_WHITESPACE	Separate whitespace at the beginning and end from the non-whitespace text.
protected Integer	seed
protected Double	temperature
static String	TEMPLATE_SINGLETRANSLATION	Template for GPTChatMessagesTemplate to translate a single word or phrase.

Constructor Summary

Constructors

Constructor	Description
GPTTranslationServiceImpl()

Method Summary

Modifier and Type	Method	Description
protected void	activate(GPTTranslationServiceImpl.Config config)
protected void	cacheResponse(String cacheKey, GPTChatRequest request, String response)
protected void	deactivate()
protected void	ensureEnabled()
protected static String	fakeTranslation(String text)	This turns the capitalization of every odd letter in each word on it's head.
protected List<String>	fragmentedTranslation(List<String> texts, String targetLanguage, GPTConfiguration configuration, AtomicInteger permittedRetries, List<GPTResponseCheck> translationChecks)
List<String>	fragmentedTranslation(List<String> texts, String targetLanguage, GPTConfiguration configuration, List<GPTResponseCheck> translationChecks)	We join all text fragments we have to translate into one big texts separated with separators like `%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 573472 %%%%%%%%%%%%%%%%` and then translate that.
protected List<String>	fragmentedTranslationDivideAndConquer(List<String> texts, String targetLanguage, GPTConfiguration configuration, AtomicInteger permittedRetries, List<GPTResponseCheck> translationChecks)	We try to translate the whole lot of texts.
protected String	getCachedResponse(String cacheKey)
protected GPTConfiguration	getServiceConfiguration()
protected static String	joinTexts(List<String> texts, List<String> ids)
protected static List<String>	separateResultTexts(String response, List<String> texts, List<String> ids, String joinedtexts)
String	singleTranslation(String rawText, String sourceLanguage, String targetLanguage, GPTConfiguration configuration)	Translate the text from the target to destination language, either Java locale name or language name.
void	streamingSingleTranslation(String text, String sourceLanguage, String targetLanguage, GPTConfiguration configuration, GPTCompletionCallback callback)	Translate the text from the target to destination language, either Java locale name or language name.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Methods inherited from interface com.composum.ai.backend.base.service.chat.GPTTranslationService

fragmentedTranslation

Field Detail

TEMPLATE_SINGLETRANSLATION

public static final String TEMPLATE_SINGLETRANSLATION

Template for GPTChatMessagesTemplate to translate a single word or phrase. Has placeholders ${sourcelanguage} ${sourcephrase} and ${targetlanguage}.

See Also:

Constant Field Values

chatCompletionService

protected GPTChatCompletionService chatCompletionService

config

protected GPTTranslationServiceImpl.Config config

temperature

protected Double temperature

seed

protected Integer seed

cacheDir

protected Path cacheDir

MULTITRANSLATION_SEPARATOR_START

public static final String MULTITRANSLATION_SEPARATOR_START

Start of separator like `%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 573472 %%%%%%%%%%%%%%%%` .

See Also:

Constant Field Values

MULTITRANSLATION_SEPARATOR_END

public static final String MULTITRANSLATION_SEPARATOR_END

End of separator like `573472 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%` .

See Also:

Constant Field Values

MULTITRANSLATION_SEPARATOR_PATTERN

protected static final Pattern MULTITRANSLATION_SEPARATOR_PATTERN

Regexp matching separator like `%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 573472 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%` (group "id" matches the number). The \n cannot be directly matched since at the start it's sometimes ```%%%%... We give the pattern a bit of leeway since some models get the number of % wrong.

LASTID

public static final String LASTID

See Also:

Constant Field Values

PATTERN_HAS_LETTER

protected static final Pattern PATTERN_HAS_LETTER

PATTERN_SEPARATE_WHITESPACE

protected static final Pattern PATTERN_SEPARATE_WHITESPACE

Separate whitespace at the beginning and end from the non-whitespace text.

HTML_TAG_AT_START

public static final Pattern HTML_TAG_AT_START

Constructor Detail

GPTTranslationServiceImpl

public GPTTranslationServiceImpl()

Method Detail

singleTranslation

@Nullable
public String singleTranslation(@Nullable
                                String rawText,
                                @Nullable
                                String sourceLanguage,
                                @Nullable
                                String targetLanguage,
                                @Nullable
                                GPTConfiguration configuration)

Translate the text from the target to destination language, either Java locale name or language name.

Specified by:

singleTranslation in interface GPTTranslationService

Parameters:

rawText - the text to translate

sourceLanguage - the language to translate from - human readable name, or null for autodetect

targetLanguage - the language to translate to - human readable name

configuration - the configuration to use

streamingSingleTranslation

public void streamingSingleTranslation(@Nonnull
                                       String text,
                                       @Nonnull
                                       String sourceLanguage,
                                       @Nonnull
                                       String targetLanguage,
                                       @Nullable
                                       GPTConfiguration configuration,
                                       @Nonnull
                                       GPTCompletionCallback callback)
                                throws GPTException

Description copied from interface: GPTTranslationService

Translate the text from the target to destination language, either Java locale name or language name.

Specified by:

streamingSingleTranslation in interface GPTTranslationService

Throws:

GPTException

fragmentedTranslation

@Nonnull
public List<String> fragmentedTranslation(@Nonnull
                                          List<String> texts,
                                          @Nonnull
                                          String targetLanguage,
                                          @Nullable
                                          GPTConfiguration configuration,
                                          @Nullable
                                          List<GPTResponseCheck> translationChecks)
                                   throws GPTException

We join all text fragments we have to translate into one big texts separated with separators like `%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 573472 %%%%%%%%%%%%%%%%` and then translate that. Then we split the result at the separators and return the fragments. Safety check is that the is from the fragments have to match the ids in the result.

Specified by:

fragmentedTranslation in interface GPTTranslationService

Parameters:

texts - the texts to translate

targetLanguage - the language to translate to - human readable name

configuration - the configuration to use

translationChecks - additional checks to verify the translation

Returns:

the translated texts

Throws:

GPTException

fragmentedTranslationDivideAndConquer

protected List<String> fragmentedTranslationDivideAndConquer(@Nonnull
                                                             List<String> texts,
                                                             @Nonnull
                                                             String targetLanguage,
                                                             @Nullable
                                                             GPTConfiguration configuration,
                                                             @Nonnull
                                                             AtomicInteger permittedRetries,
                                                             List<GPTResponseCheck> translationChecks)
                                                      throws GPTException

We try to translate the whole lot of texts. If that leads to an exception because we are out of tokens or the response was garbled, we split it into two and translate these individually. If even one text is too long, we are lost and give up.

Throws:

GPTException

fragmentedTranslation

protected List<String> fragmentedTranslation(@Nonnull
                                             List<String> texts,
                                             @Nonnull
                                             String targetLanguage,
                                             @Nullable
                                             GPTConfiguration configuration,
                                             @Nonnull
                                             AtomicInteger permittedRetries,
                                             @Nullable
                                             List<GPTResponseCheck> translationChecks)
                                      throws GPTException

Throws:

GPTException

joinTexts

protected static String joinTexts(List<String> texts,
                                  List<String> ids)

separateResultTexts

protected static List<String> separateResultTexts(String response,
                                                  List<String> texts,
                                                  List<String> ids,
                                                  String joinedtexts)

getServiceConfiguration

protected GPTConfiguration getServiceConfiguration()

fakeTranslation

protected static String fakeTranslation(String text)

This turns the capitalization of every odd letter in each word on it's head. If we are in a HTML tag (that is, between a < and a > ) then nothing is changed to avoid destroying richtext. For quick and inexpensive testing e.g. of bulk translation mechanics.

Example: "This is a test and some Code" -> "THiS iS a tEsT aNd sOmE COdE"

cacheResponse

protected void cacheResponse(String cacheKey,
                             GPTChatRequest request,
                             String response)

getCachedResponse

protected String getCachedResponse(String cacheKey)

activate

protected void activate(GPTTranslationServiceImpl.Config config)

deactivate

protected void deactivate()

ensureEnabled

protected void ensureEnabled()