lm-lib first commit;

git-svn-id: svn://home0/lm-lib@3603 71dcc7fd-b1f2-4e5d-bb8e-274146e81173
This commit is contained in:
(no author)
2024-12-16 20:58:04 +00:00
commit c1a283305d
49 changed files with 3159 additions and 0 deletions

View File

@@ -0,0 +1,65 @@
package de.extio.lmlib.client.textcompletion;
import java.time.Duration;
import java.util.function.Consumer;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import org.springframework.beans.factory.annotation.Qualifier;
import org.springframework.http.HttpMethod;
import org.springframework.web.reactive.function.client.WebClient;
import de.extio.lmlib.client.textcompletion.dto.Model;
import de.extio.lmlib.client.textcompletion.dto.ModelsResponse;
import reactor.util.retry.Retry;
class ModelNameLoader implements Runnable {
private static final Logger LOGGER = LoggerFactory.getLogger(ModelNameLoader.class);
private final WebClient webClient;
private final Consumer<String> consumer;
public ModelNameLoader(@Qualifier("textCompletionWebClientBuilder") final WebClient webClient, final Consumer<String> consumer) {
this.consumer = consumer;
this.webClient = webClient;
}
@Override
public void run() {
String modelName = null;
do {
try {
LOGGER.info("Querying model name");
modelName = this.webClient
.method(HttpMethod.GET)
.uri(uriBuilder -> uriBuilder.path("/v1/models").build())
.retrieve()
.bodyToMono(ModelsResponse.class)
.retryWhen(Retry.backoff(99, Duration.ofSeconds(1)).maxBackoff(Duration.ofSeconds(10)).jitter(0.25d).doAfterRetry(rs -> LOGGER.warn("Request failed", rs.failure())))
.block()
.getData()
.stream()
.map(Model::getId)
.findFirst()
.orElseThrow();
}
catch (final Exception e) {
LOGGER.error(e.getMessage(), e);
try {
Thread.sleep(1000l);
}
catch (final InterruptedException e1) {
throw new RuntimeException("Interrupted", e1);
}
}
} while (modelName == null);
LOGGER.info("Model name: {}", modelName);
this.consumer.accept(modelName);
}
}

View File

@@ -0,0 +1,61 @@
package de.extio.lmlib.client.textcompletion;
import java.util.Map;
import java.util.concurrent.CompletableFuture;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.CountDownLatch;
import org.springframework.beans.factory.InitializingBean;
import org.springframework.beans.factory.annotation.Autowired;
import org.springframework.beans.factory.annotation.Qualifier;
import org.springframework.stereotype.Component;
import org.springframework.web.reactive.function.client.WebClient;
import de.extio.lmlib.client.profile.ModelProfile.ModelProvider;
import de.extio.lmlib.client.profile.ModelProfileService;
@Component
class ModelNameSupplier implements InitializingBean {
@Autowired
@Qualifier("lmLibWebClientBuilder")
private WebClient.Builder webClientBuilder;
private final Map<String, CompletableFuture<String>> modelNames = new ConcurrentHashMap<>();
private final Map<String, CountDownLatch> countDownLatches = new ConcurrentHashMap<>();
@Autowired
private ModelProfileService modelProfileService;
@Override
public void afterPropertiesSet() throws Exception {
final var urls = this.modelProfileService.getModelProfileUrls(ModelProvider.OAI_TEXT_COMPLETION);
for (final var url : urls) {
this.countDownLatches.put(url, new CountDownLatch(1));
this.fetchModelName(url);
}
}
private void fetchModelName(final String url) {
final var webClient = this.webClientBuilder.baseUrl(url).build();
CompletableFuture.runAsync(() -> {
final var modelNameLoader = new ModelNameLoader(webClient, modelName -> {
this.modelNames.put(url, CompletableFuture.completedFuture(modelName));
this.countDownLatches.get(url).countDown();
});
modelNameLoader.run();
});
}
public String getModelName(final String url) {
try {
this.countDownLatches.get(url).await();
}
catch (final InterruptedException e) {
throw new RuntimeException("Interrupted", e);
}
return this.modelNames.get(url).join();
}
}

View File

@@ -0,0 +1,304 @@
package de.extio.lmlib.client.textcompletion;
import java.time.Duration;
import java.time.LocalDateTime;
import java.util.ArrayDeque;
import java.util.ArrayList;
import java.util.Deque;
import java.util.List;
import java.util.stream.Collectors;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import org.springframework.beans.factory.DisposableBean;
import org.springframework.beans.factory.annotation.Autowired;
import org.springframework.beans.factory.annotation.Qualifier;
import org.springframework.beans.factory.annotation.Value;
import org.springframework.http.HttpMethod;
import org.springframework.stereotype.Component;
import org.springframework.web.reactive.function.client.WebClient;
import de.extio.lmlib.client.Client;
import de.extio.lmlib.client.Completion;
import de.extio.lmlib.client.Conversation;
import de.extio.lmlib.client.Conversation.Turn;
import de.extio.lmlib.client.profile.ModelCategory;
import de.extio.lmlib.client.profile.ModelProfile;
import de.extio.lmlib.client.profile.ModelProfile.ModelProvider;
import de.extio.lmlib.client.profile.ModelProfileService;
import de.extio.lmlib.client.prompt.PromptStrategy;
import de.extio.lmlib.client.prompt.PromptStrategyFactory;
import de.extio.lmlib.client.textcompletion.dto.Choice;
import de.extio.lmlib.client.textcompletion.dto.CompletionRequest;
import de.extio.lmlib.client.textcompletion.dto.CompletionResponse;
import de.extio.lmlib.client.token.Tokenizer;
import reactor.util.retry.Retry;
@Component
public class TextCompletionClient implements Client, DisposableBean {
private static final Logger LOGGER = LoggerFactory.getLogger(TextCompletionClient.class);
@Autowired
@Qualifier("lmLibWebClientBuilder")
private WebClient.Builder webClientBuilder;
@Autowired
private PromptStrategyFactory promptStrategyFactory;
@Autowired
private ModelNameSupplier modelNameSupplier;
@Autowired
private Tokenizer tokenizer;
@Autowired
private ModelProfileService modelProfileService;
@Value("${client.collectStatistics:false}")
private boolean collectStatistics;
private final CompletionStatistics totalStatistics = new CompletionStatistics();
@Override
public ModelProvider getModelProvider() {
return ModelProvider.OAI_TEXT_COMPLETION;
}
@Override
public Completion completion(final ModelCategory modelCategory, final String system, final String question, final String fullText) {
String text = "";
if (!question.isEmpty() && !fullText.isEmpty()) {
text = question + "\n" + fullText;
}
else if (!question.isEmpty()) {
text = question;
}
else if (!fullText.isEmpty()) {
text = fullText;
}
else {
throw new IllegalArgumentException("Either question or fullText must be set");
}
return this.conversation(modelCategory, Conversation.create(system, text));
}
@Override
public Completion conversation(ModelCategory modelCategory, final Conversation conversation) {
if (modelCategory == null) {
modelCategory = ModelCategory.MEDIUM;
}
final var modelProfile = this.modelProfileService.getModelProfile(modelCategory.getModelProfile());
if (modelProfile == null || modelProfile.modelProvider() != ModelProvider.OAI_TEXT_COMPLETION) {
throw new IllegalArgumentException("ModelProfile " + modelCategory.getModelProfile());
}
final PromptStrategy promptStrategy = this.promptStrategyFactory.getStrategy(modelProfile.prompt());
if (promptStrategy == null) {
throw new IllegalArgumentException("Prompt strategy not found: " + modelProfile.prompt());
}
final List<StringBuilder> prompts = this.createPrompts(conversation, modelCategory, modelProfile, promptStrategy);
final List<StringBuilder> answers = new ArrayList<>();
final CompletionStatistics statistics = new CompletionStatistics();
for (final StringBuilder prompt : prompts) {
final StringBuilder answer = new StringBuilder();
this.requestCompletionsContinuations(prompt, answer, statistics, modelProfile, promptStrategy);
answers.add(answer);
}
if (answers.size() > 1) {
this.summarize(answers, modelProfile, promptStrategy, statistics);
}
return new Completion(
answers.stream().collect(Collectors.joining()),
statistics.requests,
statistics.duration,
statistics.inTokens,
statistics.outTokens,
false);
}
private List<StringBuilder> createPrompts(final Conversation conversation, final ModelCategory modelCategory, final ModelProfile modelProfile, final PromptStrategy promptStrategy) {
StringBuilder prompt = this.formatConversation(conversation, promptStrategy);
List<Long> tokenized = this.tokenizer.tokenize(prompt.toString(), modelProfile);
final int maxLength = modelProfile.maxContextLength() - modelProfile.maxTokens() - 15; // 15 is a margin for EOT and other special tokens
if (tokenized.size() > maxLength) {
// TODO: Cut smarter, e.g. split conversation into parts while keeping system prompt
LOGGER.warn("Prompt too long: {} tokens / {} max. Prompt will be cut!", tokenized.size(), maxLength);
tokenized = tokenized.subList(0, maxLength);
prompt = new StringBuilder(this.tokenizer.detokenize(tokenized, modelProfile));
}
LOGGER.info("Completion request for {}. Input tokens: {}", modelCategory, tokenized.size());
return List.of(prompt);
}
private StringBuilder formatConversation(final Conversation conversation, final PromptStrategy promptStrategy) {
final StringBuilder prompt = new StringBuilder();
final Deque<Turn> turns = new ArrayDeque<>(conversation.getConversation());
Turn systemTurn = null;
Turn userTurn = null;
Turn assistantTurn = null;
boolean first = true;
while (!turns.isEmpty()) {
final Turn turn = turns.pop();
switch (turn.type()) {
case SYSTEM:
systemTurn = turn;
break;
case USER:
userTurn = turn;
break;
case ASSISTANT:
assistantTurn = turn;
break;
default:
throw new IllegalArgumentException("Unknown turn type: " + turn.type());
}
if (first && userTurn != null) {
prompt.append(promptStrategy.start(systemTurn != null ? systemTurn.text() : "", "", userTurn.text()));
userTurn = null;
first = false;
}
else if (!first && assistantTurn != null && userTurn != null) {
promptStrategy.next(prompt, assistantTurn.text(), userTurn.text());
userTurn = null;
assistantTurn = null;
}
}
return prompt;
}
private void requestCompletionsContinuations(final StringBuilder prompt, final StringBuilder answer, final CompletionStatistics statistics,
final ModelProfile modelProfile, final PromptStrategy promptStrategy) {
for (int continuation = 0; continuation < modelProfile.maxContinuations(); continuation++) {
final LocalDateTime start = LocalDateTime.now();
final String promptStr = prompt.toString();
final var response = this.requestCompletions(promptStr, modelProfile);
String content = null;
var complete = false;
if (response.getContent() != null) {
content = promptStrategy.removeEOT(response.getContent());
complete = response.isStoppedEos();
}
else {
final var choice = response.getChoices().getFirst();
content = promptStrategy.removeEOT(choice.getText());
complete = !Choice.FINISH_REASON_LENGTH.equals(choice.getFinishReason());
}
if (this.collectStatistics) {
final var dur = Duration.between(start, LocalDateTime.now());
final var in = this.tokenizer.count(promptStr, modelProfile);
final var out = this.tokenizer.count(content, modelProfile);
LOGGER.debug("Request duration: " + dur + "; in tokens: " + in + "; out tokens: " + out);
statistics.add(dur, in, out);
this.totalStatistics.add(dur, in, out);
}
answer.append(content);
if (complete) {
LOGGER.debug("Prompt response complete");
break;
}
answer.append('\n');
promptStrategy.continue_(prompt, content);
}
}
private CompletionResponse requestCompletions(final String prompt, final ModelProfile modelProfile) {
LOGGER.debug("Requesting completion at {}", modelProfile.url());
final var request = new CompletionRequest();
request.setModel(this.modelNameSupplier.getModelName(modelProfile.url()));
request.setPrompt(prompt);
request.setMaxTokens(modelProfile.maxTokens());
request.setTemperature(modelProfile.temperature());
request.setTopP(modelProfile.topP());
request.setStream(false);
final var webClient = this.webClientBuilder.baseUrl(modelProfile.url()).build();
return webClient
.method(HttpMethod.POST)
.uri(uriBuilder -> uriBuilder.path("/v1/completions").build())
.bodyValue(request)
.retrieve()
.bodyToMono(CompletionResponse.class)
.retryWhen(Retry.backoff(99, Duration.ofSeconds(1)).maxBackoff(Duration.ofSeconds(10)).jitter(0.25d).doAfterRetry(rs -> LOGGER.warn("Request failed", rs.failure())))
.block();
}
private void summarize(final List<StringBuilder> answers, final ModelProfile modelProfile, final PromptStrategy promptStrategy, final CompletionStatistics statistics) {
// TODO: Reactivate when multiple (split) prompts have been implemented again
// LOGGER.debug("Summarizing answers");
//
// final StringBuilder summary = new StringBuilder();
//
// final StringBuilder prompt = promptStrategy.start(instruction, question, "");
// answers.forEach(answer -> promptStrategy.next(prompt, answer.toString(), "Continue"));
// promptStrategy.next(prompt, "", "Generate now a full summary");
//
// this.requestCompletionsContinuations(prompt, summary, statistics, modelProfile, promptStrategy);
//
// answers.clear();
// answers.add(summary);
}
@Override
public void destroy() throws Exception {
if (this.collectStatistics) {
LOGGER.info("Client Statistics: {}", this.totalStatistics);
}
}
static class CompletionStatistics {
volatile int requests;
volatile Duration duration = Duration.ofMillis(0l);
volatile long inTokens;
volatile long outTokens;
synchronized void add(final Duration duration, final int inTokens, final int outTokens) {
this.requests++;
this.duration = this.duration.plus(duration);
this.inTokens += inTokens;
this.outTokens += outTokens;
}
@Override
public String toString() {
final StringBuilder builder = new StringBuilder();
builder.append("CompletionStatistics [requests=");
builder.append(this.requests);
builder.append(", duration=");
builder.append(this.duration);
builder.append(", inTokens=");
builder.append(this.inTokens);
builder.append(", outTokens=");
builder.append(this.outTokens);
builder.append("]");
return builder.toString();
}
}
}

View File

@@ -0,0 +1,59 @@
package de.extio.lmlib.client.textcompletion.dto;
import com.fasterxml.jackson.annotation.JsonProperty;
public final class Choice {
public static final String FINISH_REASON_STOP = "stop";
public static final String FINISH_REASON_LENGTH = "length";
public static final String FINISH_REASON_CONTENT_FILTER = "content_filter";
private String text;
private Double score;
private String finishReason;
@JsonProperty("text")
public String getText() {
return this.text;
}
public void setText(final String text) {
this.text = text;
}
@JsonProperty("score")
public Double getScore() {
return this.score;
}
public void setScore(final Double score) {
this.score = score;
}
@JsonProperty("finish_reason")
public String getFinishReason() {
return this.finishReason;
}
public void setFinishReason(final String finishReason) {
this.finishReason = finishReason;
}
@Override
public String toString() {
final StringBuilder builder = new StringBuilder();
builder.append("Choice [text=");
builder.append(this.text);
builder.append(", score=");
builder.append(this.score);
builder.append(", finishReason=");
builder.append(this.finishReason);
builder.append("]");
return builder.toString();
}
}

View File

@@ -0,0 +1,72 @@
package de.extio.lmlib.client.textcompletion.dto;
import com.fasterxml.jackson.annotation.JsonProperty;
public final class CompletionRequest {
private String model;
private String prompt;
private Integer maxTokens;
private Double temperature;
private Double topP;
private Boolean stream;
@JsonProperty("model")
public String getModel() {
return this.model;
}
public void setModel(final String model) {
this.model = model;
}
@JsonProperty("prompt")
public String getPrompt() {
return this.prompt;
}
public void setPrompt(final String prompt) {
this.prompt = prompt;
}
@JsonProperty("max_tokens")
public Integer getMaxTokens() {
return this.maxTokens;
}
public void setMaxTokens(final Integer maxTokens) {
this.maxTokens = maxTokens;
}
@JsonProperty("temperature")
public Double getTemperature() {
return this.temperature;
}
public void setTemperature(final Double temperature) {
this.temperature = temperature;
}
@JsonProperty("top_p")
public Double getTopP() {
return this.topP;
}
public void setTopP(final Double topP) {
this.topP = topP;
}
@JsonProperty("stream")
public Boolean isStream() {
return this.stream;
}
public void setStream(final Boolean stream) {
this.stream = stream;
}
}

View File

@@ -0,0 +1,93 @@
package de.extio.lmlib.client.textcompletion.dto;
import java.util.List;
import com.fasterxml.jackson.annotation.JsonProperty;
public final class CompletionResponse {
private String id;
private String object;
private Integer created;
private List<Choice> choices;
private String content;
private boolean isStoppedEos;
@JsonProperty("id")
public String getId() {
return this.id;
}
public void setId(final String id) {
this.id = id;
}
@JsonProperty("object")
public String getObject() {
return this.object;
}
public void setObject(final String object) {
this.object = object;
}
@JsonProperty("created")
public Integer getCreated() {
return this.created;
}
public void setCreated(final Integer created) {
this.created = created;
}
@JsonProperty("choices")
public List<Choice> getChoices() {
return this.choices;
}
public void setChoices(final List<Choice> choices) {
this.choices = choices;
}
@JsonProperty("content")
public String getContent() {
return this.content;
}
public void setContent(final String content) {
this.content = content;
}
@JsonProperty("stopped_eos")
public boolean isStoppedEos() {
return this.isStoppedEos;
}
public void setStoppedEos(final boolean isStoppedEos) {
this.isStoppedEos = isStoppedEos;
}
@Override
public String toString() {
final StringBuilder builder = new StringBuilder();
builder.append("CompletionResponse [id=");
builder.append(this.id);
builder.append(", object=");
builder.append(this.object);
builder.append(", created=");
builder.append(this.created);
builder.append(", choices=");
builder.append(this.choices);
builder.append(", content=");
builder.append(this.content);
builder.append(", isStoppedEos=");
builder.append(this.isStoppedEos);
builder.append("]");
return builder.toString();
}
}

View File

@@ -0,0 +1,74 @@
package de.extio.lmlib.client.textcompletion.dto;
import java.util.List;
import com.fasterxml.jackson.annotation.JsonProperty;
public final class Model {
private String id;
private String object;
private int created;
private String ownedBy;
private List<String> permission;
private String root;
@JsonProperty("id")
public String getId() {
return this.id;
}
public void setId(final String id) {
this.id = id;
}
@JsonProperty("object")
public String getObject() {
return this.object;
}
public void setObject(final String object) {
this.object = object;
}
@JsonProperty("created")
public int getCreated() {
return this.created;
}
public void setCreated(final int created) {
this.created = created;
}
@JsonProperty("owned_by")
public String getOwned_by() {
return this.ownedBy;
}
public void setOwned_by(final String owned_by) {
this.ownedBy = owned_by;
}
@JsonProperty("permission")
public List<String> getPermission() {
return this.permission;
}
public void setPermission(final List<String> permission) {
this.permission = permission;
}
@JsonProperty("root")
public String getRoot() {
return this.root;
}
public void setRoot(final String root) {
this.root = root;
}
}

View File

@@ -0,0 +1,42 @@
package de.extio.lmlib.client.textcompletion.dto;
import java.util.List;
import com.fasterxml.jackson.annotation.JsonProperty;
public final class ModelsResponse {
private String object;
private List<Model> data;
@JsonProperty("object")
public String getObject() {
return this.object;
}
public void setObject(final String object) {
this.object = object;
}
@JsonProperty("data")
public List<Model> getData() {
return this.data;
}
public void setData(final List<Model> data) {
this.data = data;
}
@Override
public String toString() {
final StringBuilder builder = new StringBuilder();
builder.append("ModelsResponse [object=");
builder.append(this.object);
builder.append(", data=");
builder.append(this.data);
builder.append("]");
return builder.toString();
}
}