Having issues with qdrant vector store when adding document

[obaume]

I want to create a chatbot that answers questions based on a knowledge base. The knowledge base is stored in the question model.

I made an Artisan command to test embedding and storing the knowledge base in the vector store.

I used the data loading example from the documentation.

I had a problem with the document ids, so I added them manually.

<?php

namespace App\Console\Commands;

use App\Models\Question;
use Illuminate\Console\Command;
use NeuronAI\RAG\DataLoader\StringDataLoader;
use NeuronAI\RAG\Embeddings\OllamaEmbeddingsProvider;
use NeuronAI\RAG\VectorStore\QdrantVectorStore;

class CreateEmbeddingsForDB extends Command
{
    ...
    /**
     * Execute the console command.
     */
    public function handle()
    {
        $questions = Question::all();

        $provider = new OllamaEmbeddingsProvider(
            url:'HOSTED_OLLAMA_INSTANCE',
            model:'nomic-embed-text:latest'
        );

        $store = new QdrantVectorStore(
            collectionUrl: 'http://localhost:6333/collections/cs_site_rag/',
            key:'',
        );

        $bar = $this->output->createProgressBar(count($questions));

        foreach ($questions as $question) {
            $documents = StringDataLoader::for(
               $question->question . " " . strip_tags($question->answer)
             )->getDocuments();

            foreach ($documents as $document) {
                $document->id = $question->id;
            }

            $embededDocuments = $provider->embedDocuments($documents);

            $store->addDocuments($embededDocuments);
        }
    }
}

When i run the command i get this error message :

Client error: PUT http://localhost:6333/collections/cs_site_rag/points resulted in a 400 Bad Request response:
{"status":{"error":"Format error in JSON body: Invalid PointInsertOperations format"},"time":0.0}

After looking in the NeuronAI and QDrant docs, I couldn't find an explanation. What could be the cause of this error ?

[ilvalerione]

You could try the latest version 1.9.23 that added batch operations for Qdrant. Let me know if it works for you.

[obaume]

When i tried with version 1.9.24 i got this new error :

Client error: `PUT http://localhost:6333/collections/cs_site_rag/points/batch` resulted in a `404 Not Found` `response

I also tried using addDocument rather than addDocuments and it worked. It didn't work with verison 1.9.20.

Unfortunetly when i try to use the similarity search i got this error from QdrantVectorStore Undefined array key "vector". Which i believe is due to the fact that the reponse from the client->post looks like this :

{
   "result":[
      {
         "id":142,
         "version":429,
         "score":0.8458562
      },
      {
         "id":1,
         "version":283,
         "score":0.8458562
      },
      {
         "id":145,
         "version":432,
         "score":0.72743607
      },
      {
         "id":4,
         "version":286,
         "score":0.72743607
      }
   ],
   "status":"ok",
   "time":0.003649385
}

Nonetheless thank you for the quick response!

[sixty-nine]

Hi, I'm having the same problems.

Adding documents in batch does not work, using addDocument does.

Then there is the problem with the similarity search which I was able to fix by adding the with_payload and with_vector parameters to the request:

        // In: QdrantVectorStore::similaritySearch
        $response = $this->client->post('points/search', [
            RequestOptions::JSON => [
                'vector' => $embedding,
                'limit' => $this->topK,
                'with_payload' => true,  // <------------------------ HERE
                'with_vector' => true,  // <------------------------ AND HERE

            ]
        ])->getBody()->getContents();

[ilvalerione]

Thank you for your report. I included this fix in the 1.9.25 and also fixed the batch upload. It was implemented following this doc: https://qdrant.tech/documentation/concepts/points/#batch-update

Unfortunately I didn't find a way to include Qdrant in the Neuron github action to perform tests like Elasticsearch and Typesense. If someone know how to do a PR is welcomed.

Let me know if it works for you now.