Skip to content

Google Vertex AI toolcalling token usage #1963

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 1 commit into from
Closed

Google Vertex AI toolcalling token usage #1963

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
77 changes: 57 additions & 20 deletions ...-gemini/src/main/java/org/springframework/ai/vertexai/gemini/VertexAiGeminiChatModel.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,7 +47,6 @@
import io.micrometer.observation.Observation;
import io.micrometer.observation.ObservationRegistry;
import io.micrometer.observation.contextpropagation.ObservationThreadLocalAccessor;
import org.springframework.ai.vertexai.gemini.common.VertexAiGeminiSafetySetting;
import reactor.core.publisher.Flux;

import org.springframework.ai.chat.messages.AssistantMessage;
Expand All @@ -58,6 +57,9 @@
import org.springframework.ai.chat.messages.UserMessage;
import org.springframework.ai.chat.metadata.ChatGenerationMetadata;
import org.springframework.ai.chat.metadata.ChatResponseMetadata;
import org.springframework.ai.chat.metadata.EmptyUsage;
import org.springframework.ai.chat.metadata.Usage;
import org.springframework.ai.chat.metadata.UsageUtils;
import org.springframework.ai.chat.model.AbstractToolCallSupport;
import org.springframework.ai.chat.model.ChatModel;
import org.springframework.ai.chat.model.ChatResponse;
Expand All @@ -77,6 +79,7 @@
import org.springframework.ai.model.function.FunctionCallingOptions;
import org.springframework.ai.retry.RetryUtils;
import org.springframework.ai.vertexai.gemini.common.VertexAiGeminiConstants;
import org.springframework.ai.vertexai.gemini.common.VertexAiGeminiSafetySetting;
import org.springframework.ai.vertexai.gemini.metadata.VertexAiUsage;
import org.springframework.beans.factory.DisposableBean;
import org.springframework.lang.NonNull;
Expand All @@ -95,6 +98,7 @@
* @author Mark Pollack
* @author Soby Chacko
* @author Jihoon Kim
* @author Ilayaperumal Gopinathan
* @since 0.8.1
*/
public class VertexAiGeminiChatModel extends AbstractToolCallSupport implements ChatModel, DisposableBean {
Expand Down Expand Up @@ -286,6 +290,10 @@ private static Schema jsonToSchema(String json) {
// https://cloud.google.com/vertex-ai/docs/generative-ai/model-reference/gemini
@Override
public ChatResponse call(Prompt prompt) {
return this.internalCall(prompt, null);
}

public ChatResponse internalCall(Prompt prompt, ChatResponse previousChatResponse) {

VertexAiGeminiChatOptions vertexAiGeminiChatOptions = vertexAiGeminiChatOptions(prompt);

Expand All @@ -310,8 +318,10 @@ public ChatResponse call(Prompt prompt) {
.flatMap(List::stream)
.toList();

ChatResponse chatResponse = new ChatResponse(generations,
toChatResponseMetadata(generateContentResponse));
GenerateContentResponse.UsageMetadata usage = generateContentResponse.getUsageMetadata();
Usage currentUsage = (usage != null) ? new VertexAiUsage(usage) : new EmptyUsage();
Usage cumulativeUsage = UsageUtils.getCumulativeUsage(currentUsage, previousChatResponse);
ChatResponse chatResponse = new ChatResponse(generations, toChatResponseMetadata(cumulativeUsage));

observationContext.setResponse(chatResponse);
return chatResponse;
Expand All @@ -321,7 +331,7 @@ public ChatResponse call(Prompt prompt) {
var toolCallConversation = handleToolCalls(prompt, response);
// Recursively call the call method with the tool call message
// conversation that contains the call responses.
return this.call(new Prompt(toolCallConversation, prompt.getOptions()));
return this.internalCall(new Prompt(toolCallConversation, prompt.getOptions()), response);
}

return response;
Expand All @@ -330,6 +340,10 @@ public ChatResponse call(Prompt prompt) {

@Override
public Flux<ChatResponse> stream(Prompt prompt) {
return this.internalStream(prompt, null);
}

public Flux<ChatResponse> internalStream(Prompt prompt, ChatResponse previousChatResponse) {
return Flux.deferContextual(contextView -> {
VertexAiGeminiChatOptions vertexAiGeminiChatOptions = vertexAiGeminiChatOptions(prompt);

Expand All @@ -349,32 +363,51 @@ public Flux<ChatResponse> stream(Prompt prompt) {
try {
ResponseStream<GenerateContentResponse> responseStream = request.model
.generateContentStream(request.contents);

return Flux.fromStream(responseStream.stream()).switchMap(response -> {

Flux<ChatResponse> chatResponseFlux = Flux.fromStream(responseStream.stream()).switchMap(response -> {
List<Generation> generations = response.getCandidatesList()
.stream()
.map(this::responseCandidateToGeneration)
.flatMap(List::stream)
.toList();

ChatResponse chatResponse = new ChatResponse(generations, toChatResponseMetadata(response));

if (!isProxyToolCalls(prompt, this.defaultOptions) && isToolCall(chatResponse,
GenerateContentResponse.UsageMetadata usage = response.getUsageMetadata();
Usage currentUsage = (usage != null) ? new VertexAiUsage(usage) : new EmptyUsage();
Usage cumulativeUsage = UsageUtils.getCumulativeUsage(currentUsage, previousChatResponse);
ChatResponse chatResponse = new ChatResponse(generations, toChatResponseMetadata(cumulativeUsage));
return Flux.just(chatResponse);
}).buffer(2, 1).filter(bufferedResponses -> {
return bufferedResponses.size() == 2;
}).map(bufferList -> {
ChatResponse firstResponse = bufferList.get(0);
ChatResponse secondResponse = bufferList.get(1);
if (secondResponse != null && secondResponse.getMetadata() != null) {
// This is the usage from the final Chat response for a
// given Chat request.
Usage usage = secondResponse.getMetadata().getUsage();
if (!UsageUtils.isEmpty(usage)) {
// Store the usage from the final response to the
// penultimate response for accumulation.
return new ChatResponse(firstResponse.getResults(), toChatResponseMetadata(usage));
}
}
return firstResponse;
});
Flux<ChatResponse> flux = chatResponseFlux.flatMap(response -> {
if (!isProxyToolCalls(prompt, this.defaultOptions) && isToolCall(response,
Set.of(FinishReason.STOP.name(), FinishReason.FINISH_REASON_UNSPECIFIED.name()))) {
var toolCallConversation = handleToolCalls(prompt, chatResponse);
var toolCallConversation = handleToolCalls(prompt, response);
// Recursively call the stream method with the tool call message
// conversation that contains the call responses.
return this.stream(new Prompt(toolCallConversation, prompt.getOptions()));
return this.internalStream(new Prompt(toolCallConversation, prompt.getOptions()), response);
}

Flux<ChatResponse> chatResponseFlux = Flux.just(chatResponse)
.doOnError(observation::error)
.doFinally(s -> observation.stop())
.contextWrite(ctx -> ctx.put(ObservationThreadLocalAccessor.KEY, observation));

return new MessageAggregator().aggregate(chatResponseFlux, observationContext::setResponse);
});
else {
return Flux.just(response);
}
})
.doOnError(observation::error)
.doFinally(s -> observation.stop())
.contextWrite(ctx -> ctx.put(ObservationThreadLocalAccessor.KEY, observation));
return new MessageAggregator().aggregate(flux, observationContext::setResponse);
}
catch (Exception e) {
throw new RuntimeException("Failed to generate content", e);
Expand Down Expand Up @@ -430,6 +463,10 @@ private ChatResponseMetadata toChatResponseMetadata(GenerateContentResponse resp
return ChatResponseMetadata.builder().withUsage(new VertexAiUsage(response.getUsageMetadata())).build();
}

private ChatResponseMetadata toChatResponseMetadata(Usage usage) {
return ChatResponseMetadata.builder().withUsage(usage).build();
}

private VertexAiGeminiChatOptions vertexAiGeminiChatOptions(Prompt prompt) {
VertexAiGeminiChatOptions updatedRuntimeOptions = VertexAiGeminiChatOptions.builder().build();
if (prompt.getOptions() != null) {
Expand Down
37 changes: 37 additions & 0 deletions ...springframework/ai/vertexai/gemini/function/VertexAiGeminiChatModelFunctionCallingIT.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -127,8 +127,14 @@ public void functionCallTestInferredOpenApiSchema() {

logger.info("Response: {}", response);

assertThat(response.getResult()).isNotNull();
assertThat(response.getResult().getOutput()).isNotNull();
assertThat(response.getResult().getOutput().getText()).containsAnyOf("15.0", "15");

assertThat(response.getMetadata()).isNotNull();
assertThat(response.getMetadata().getUsage()).isNotNull();
assertThat(response.getMetadata().getUsage().getTotalTokens()).isGreaterThan(150).isLessThan(230);

ChatResponse response2 = this.chatModel
.call(new Prompt("What is the payment status for transaction 696?", promptOptions));

Expand Down Expand Up @@ -214,6 +220,37 @@ public void functionCallTestInferredOpenApiSchemaStream() {

}

@Test
public void functionCallTestUsageInferredOpenApiSchemaStream() {

UserMessage userMessage = new UserMessage(
"What's the weather like in San Francisco, Paris and in Tokyo? Return the temperature in Celsius.");

List<Message> messages = new ArrayList<>(List.of(userMessage));

var promptOptions = VertexAiGeminiChatOptions.builder()
.withModel(VertexAiGeminiChatModel.ChatModel.GEMINI_1_5_FLASH)
.withFunctionCallbacks(List.of(FunctionCallback.builder()
.function("getCurrentWeather", new MockWeatherService())
.schemaType(SchemaType.OPEN_API_SCHEMA)
.description("Get the current weather in a given location")
.inputType(MockWeatherService.Request.class)
.build()))
.build();

Flux<ChatResponse> responseFlux = this.chatModel.stream(new Prompt(messages, promptOptions));

ChatResponse chatResponse = responseFlux.blockLast();

logger.info("Response: {}", chatResponse);

assertThat(chatResponse).isNotNull();
assertThat(chatResponse.getMetadata()).isNotNull();
assertThat(chatResponse.getMetadata().getUsage()).isNotNull();
assertThat(chatResponse.getMetadata().getUsage().getTotalTokens()).isGreaterThan(150).isLessThan(230);

}

public record PaymentInfoRequest(String id) {

}
Expand Down
Loading