When using the new .add method, unable to provide collection config

[bnkc]

Here is my example:

from qdrant_client.models import Distance, VectorParams
from qdrant_client import QdrantClient


# Initialize the client
client = QdrantClient(":memory:")  


client.create_collection(
    collection_name="test_collection",
    vectors_config=VectorParams(size=384, distance=Distance.COSINE),
)


# Prepare your documents, metadata, and IDs
docs = [
    "Qdrant has Langchain integrations",
    "Qdrant also has Llama Index integrations",
] * 5
metadata = [
    {"source": "Langchain-docs"},
    {"source": "Linkedin-docs"},
] * 5


client.add(collection_name="test_collection", documents=docs, metadata=metadata)

If I try to create_collection before applying the client.add, I get the following traceback:

Traceback (most recent call last):
  File "/Users/lev/Developer/poetry-demo/poetry_demo/qdrant.py", line 40, in <module>
    client.add(collection_name="test_collection", documents=docs, metadata=metadata)
  File "/Users/lev/.pyenv/versions/3.11.1/envs/testing-venv/lib/python3.11/site-packages/qdrant_client/qdrant_fastembed.py", line 496, in add
    self._validate_collection_info(collection_info)
  File "/Users/lev/.pyenv/versions/3.11.1/envs/testing-venv/lib/python3.11/site-packages/qdrant_client/qdrant_fastembed.py", line 348, in _validate_collection_info
    assert isinstance(
AssertionError: Collection have incompatible vector params: size=384 distance=<Distance.COSINE: 'Cosine'> hnsw_config=None quantization_config=None on_disk=None

This is an issue because if I want to use the batch_search method later on rather than the batch_query method, I can't because of a mismatching config. The reason I would rather use the batch_search is because I have already generated embedding for a large volume of text. It makes no sense to regenerate upon query. I would rather query back the original embedding and provide as a parameter to the one of the search methods:

search_result = client.search(
    collection_name="demo_collection", query_vector=[0.1, 0.2, 0.3, 0.4]
)

[joein]

Hello @bnkc

With pure qdrant-client API (without fastembed mixin's methods like add and query), you can have unnamed vectors, which means that you don't need to provide a name.

However, fastembed creates named vectors and operates with them. Names of the vectors are generated from the chosen model names.
Thus, to do search, you need to set query_vector to a some named vector, e.g.:

models.NamedVector(
    name=client.get_vector_field_name(), vector=query_vector
)

If collection was not created, then the first call to add creates it.
If you want to create it on your own, take a look at get_fastembed_vector_params method

[bnkc]

Hi @joein

Thanks! This worked for me :)

[john-pome]

I've spent a few hours trying to figure out how to get the add and query methods working. The example code I've found for these methods doesn't really seem to work in my case, except the most trivial in memory case with only three text docs in a string, and they don't seem to be officially documented. My code is the same as above except without the config and I'm getting the same error.

I don't understand how to get this answer to work in my use case, which is similar to the original poster's. If I need to learn how the underlying methods work in detail (eg what is a named vs unnamed vector), it might be helpful to update some of the public documentation since these are supposed to be "high level" methods. For instance in this answer, I don't know where models. comes from- what is it? how do I import it? or do I need to apply that when I add or when I query? Just some thoughts.

If anyone has some more info I'd appreciate it while I dig through the docs.

[john-pome]

If anyone finds this with the same problem, the key for me was that the client.add() method did not properly create the collection. I had to separately create it so that I could use this specific config method: client.get_fastembed_vector_params(). Thanks to joein for the hint above.

If it helps anyone else here's a snippet for setting up a collection.

    def _setup_collection(self):
        try:
            if not self._client.collection_exists(self._collection_name):
                logger.info(f"Collection not found, creating collection: {self._collection_name}")
                self._client.create_collection(
                    collection_name=self._collection_name,
                    vectors_config=self._client.get_fastembed_vector_params(),
                )
        except Exception as e:
            raise RuntimeError(
                f"Unable to setup collection: {self._collection_name}, Error: {e}"
            )