Quarkus LangChain4j Vertex Gemini AI Embedding does not have text-embedding-004:batchEmbedContents or gemini-embedding-001:batchEmbedContents

[semekita]

I am using the following dependency:

        <dependency>
            <groupId>io.quarkiverse.langchain4j</groupId>
            <artifactId>quarkus-langchain4j-vertex-ai-gemini</artifactId>
        </dependency> 

Here is some portion of my config:

quarkus.langchain4j.vertexai.gemini.embedding-model.model-id=gemini-embedding-001
quarkus.langchain4j.vertexai.gemini.embedding-model.log-requests=true
quarkus.langchain4j.vertexai.gemini.embedding-model.log-responses=true
quarkus.langchain4j.vertexai.gemini.embedding-model.timeout=300s
quarkus.langchain4j.vertexai.gemini.embedding-model.task-type=QUESTION_ANSWERING
quarkus.langchain4j.vertexai.gemini.base-url=https://us-central1-aiplatform.googleapis.com
quarkus.langchain4j.vertexai.gemini.location=us-central1
quarkus.langchain4j.vertexai.gemini.chat-model.model-id=gemini-2.5-pro
quarkus.langchain4j.vertexai.gemini.project-id=${GOOGLE_PROJECT_ID:myproject}

when start my code for embedding, I am getting following error below:

• method: POST
• url: https://us-central1-aiplatform.googleapis.com/v1/projects/myproject/locations/us-central1/publishers/google/models/text-embedding-004:batchEmbedContents

• The requested URL /v1/projects/myproject/locations/us-central1/publishers/google/models/text-embedding-004:batchEmbedContents was not found on this server. That?s all we know.

It does not matter what model I use, I am getting 404. Checking google documentation gemini-embedding-001 or text-embedding-004 don't have batchEmbedContents for embdding.

Please advise

[cescoffier]

\CC @geoand

[geoand]

Perhaps you can try quarkus-langchain4j-ai-gemini instead

[semekita]

I can't use quarkus-langchain4j-ai-gemini since that is not for enterprise. Did anyone can test or use quarkus-langchain4j-vertex-ai-gemini? There is quarkus-langchain4j-vertex-ai in quarkus it has right api calls but it does not have embedding model.

[geoand]

Can you show me what the code you are using looks like?

[sberyozkin]

Might be a restriction related to the

quarkus.langchain4j.vertexai.gemini.location=us-central1
quarkus.langchain4j.vertexai.gemini.chat-model.model-id=gemini-2.5-pro

combination, the location or model

[semekita]

Here is the class in quarkus being used to make call to google:
https://github.com/quarkiverse/quarkus-langchain4j/blob/main/model-providers/google/vertex-ai-gemini/runtime/src/main/java/io/quarkiverse/langchain4j/vertexai/runtime/gemini/VertxAiGeminiRestApi.java

[semekita]

Here is the code.

@ApplicationScoped
public class DocumentIngestionService {

   
    public void ingest(@Observes StartupEvent ev,
                       EmbeddingStore store, EmbeddingModel embeddingModel,
                       @ConfigProperty(name = "rag.location") Path documents) {

        System.out.print("Model >>>>>>>>>>>>>>>>> " + embeddingModel.toString());
        store.removeAll(); // cleanup the store to start fresh (just for demo purposes)
        List<Document> list = FileSystemDocumentLoader.loadDocumentsRecursively(documents);

        ingestDocument(list, store, embeddingModel);
       
        Log.info("Documents ingested successfully");
    }
    private void ingestDocument(List<Document> list, EmbeddingStore store, EmbeddingModel embeddingModel) {
         EmbeddingStoreIngestor ingestor = EmbeddingStoreIngestor.builder()
                .embeddingStore(store)
                .embeddingModel(embeddingModel)
                .documentSplitter(recursive(100, 25))
                .build();
        ingestor.ingest(list);
    }
}

Here is the link to quarkus code being used by the library https://github.com/quarkiverse/quarkus-langchain4j/blob/main/model-providers/google/vertex-ai-gemini/runtime/src/main/java/io/quarkiverse/langchain4j/vertexai/runtime/gemini/VertxAiGeminiRestApi.java

[geoand]

@sberyozkin @jmartisk I guess we shouldn't embedAll?

[semekita]

That was my reaction too it should not be embedAll.

for vertex gemini ai doc is here:
https://docs.cloud.google.com/vertex-ai/generative-ai/docs/embeddings/get-text-embeddings#googlegenaisdk_embeddings_docretrieval_with_txt-drest

for Gemini ai:
https://ai.google.dev/api/embeddings#v1beta.models.batchEmbedContents

[semekita]

@cescoffier @sberyozkin @geoand @jmartisk

Do you know the proper way to use vertex gemini ai since I can’t use gemini ai it is not for enterprise.
I would really appreciate if someone can help me.
I tried different properties but it didn’t work.
Location and project id are required.

[semekita]

@sberyozkin I try curl

curl -v -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" -H "Content-Type: application/json" https://us-east4-aiplatform.googleapis.com/v1/projects/(ProjectId)/locations/us-east4/publishers/google/models/gemini-embedding-001:predict -d $'{
  "instances": [
    { "content": ""What is Quarkus?

Here is the abbreviated json result

{
  "predictions": [
    {
      "embeddings": {
        "statistics": {
          "truncated": false,
          "token_count": 5
        },
        "values": [
          -0.015185040421783924,
          -0.010913979262113571,
          -0.00273782922886312,
          -0.011670076288282871,
          -0.033775597810745239,
          0.023877182975411415,
          0.043472196906805038,
          -0.023842435330152512,
          0.0010244137374684215,
          0.0039118132553994656,
          0.010617666877806187,
          -0.012548168189823627,
          -0.015438683331012726,
          0.016605859622359276,
          -0.0041220821440219879,
          0.021938646212220192,
          0.0083180330693721771,
          -0.012503384612500668,
          0.021779298782348633,
          0.0026342230848968029,
          -0.018437176942825317,
          0.0014919590903446078,
          -0.013023228384554386,
          0.027069881558418274,
          -0.0070800068788230419,
          -0.0050653186626732349,
          -0.014653032645583153,
          -0.0082474872469902039,
          0.0099378852173686028
        ]
      }
    }
  ],
  "metadata": {
    "billableCharacterCount": 14
  }
}

[sberyozkin]

@semekita Thanks, so, as far as https://github.com/quarkiverse/quarkus-langchain4j/blob/1.7.4/model-providers/google/vertex-ai-gemini/runtime/src/main/java/io/quarkiverse/langchain4j/vertexai/runtime/gemini/VertxAiGeminiRestApi.java is concerned, we don't want both embedContent and batchContent and only predictContent that would produce a request URL like you shown ending with ${embedding-model}:predict ?

@geoand Hi Georgios, perhaps we can move this discussion to Quarkus LangChain4j ?

I can update the Vertex AI Gemini embedding code, I guess embed will translate to that URI with :predict and embedAll should throw an unsupported operation exception...

This is clearly different from AiGemini embedding model support, so I'm not sure, :predict does not sound like it is similar to :embed, we probably need someone else with the Vertex Gemini experience to weigh in.

FYI, I don't see anything related in https://github.com/langchain4j/langchain4j/tree/main/langchain4j-vertex-ai-gemini/src/main/java/dev/langchain4j/model/vertexai/gemini

[geoand]

Hi Georgios, perhaps we can move this discussion to Quarkus LangChain4j ?

I tried but I don't seem to be able to. Maybe @gastaldi can?

[gastaldi]

Looks like discussion transfers only work inside the same org, maybe ir's easier to start a new one there

[geoand]

maybe ir's easier to start a new one there

Very likely so