serverStreamingPredict, Error : Missing field messages in input

[ryuzaki01]

https://github.com/googleapis/google-cloud-node/blob/38140975dcfc5936032d05083eb3f7b9bfe93a7c/packages/google-cloud-aiplatform/src/v1beta1/prediction_service_client.ts#L924C32-L924C38

Is this datatype actually array of number ? data type is miss match and no explantation on how to generate a Tensor Input

I have tried this

const predictClient = new PredictionServiceClient({
  apiEndpoint: 'us-central1-aiplatform.googleapis.com',
  auth
});

const endpoint `projects/.......`
const Input = {
  "context": `Your are an Ai`,
  "examples": [],
  "messages": [
    {
      "author": "user",
      "content": 'This is question'
    },
  ]
}

const parameter = {
    candidateCount: 1,
    maxOutputTokens: 1024,
    temperature: 0.2,
    topP: 0.8,
    topK: 40
};
    
 const request = {
      endpoint,
      inputs: [Tensor.fromObject(prompt)],
      parameter: Tensor.fromObject(parameter)
};
  
const stream = predictClient.serverStreamingPredict(request);
stream.on('data', (response) => { console.log(response) });
stream.on('error', (err) => { throw(err) });
stream.on('end', () => { /* API call completed */ });

it fails with message

 details: 'Missing field messages in input.',

[ryuzaki01]

Tried this

 const input = Tensor.fromObject({
      structVal: {
        context: {
          stringVal: [prompt as string]
        },
        examples: {
          listVal: []
        },
        messages: {
          listVal: [
            {
              author: {
                stringVal: ["user"]
              },
              content: {
                stringVal: [question as string]
              }
            },
          ]
        }
      }
    })
    const streamRequest = {
      endpoint,
      inputs: [input],
      parameters: parameter as ITensor
    };
    
    const stream = predictClient.serverStreamingPredict(streamRequest);
    

Still doesnt work, with the same message

 details: 'Missing field messages in input.',

NOTE: using this kind of object structure successfully created an ITensor with all value intact, last time it returned Empty Tensor object

I don't know if this is a Bug or I'm implementing it the wrong way, please give me guidance

[rgrzywinski]

So close, dude! From the REST API you can see that you're missing a structVal within your listVal.

[rgrzywinski]

The real question is: are there helpers to create ITensors like there are with IValue?

(FYI -- you don't need the Tensor.fromObject. The object you get back is exactly the same as the one you put in.)