[Question]: how to qurey based on particle embedding in database

[axz91]

Question Validation

• I have searched both the documentation and discord for an answer.

Question

collection.get(
ids=["id1", "id2", "id3", ...],
where={"style": "style1"}
)

https://docs.trychroma.com/usage-guide#querying-a-collection

got the particular embedding then do LLM query , how to do this?

[dosubot[bot]]

🤖

Hello,

To query based on a particular embedding in the LlamaIndex database after obtaining it from a collection using LLM, you can use the query method provided in the MilvusVectorStore class. This method takes a VectorStoreQuery object as input and returns a VectorStoreQueryResult object which contains the most similar nodes, their similarities, and their ids.

Here is an example of how you can use it:

from llama_index.vector_stores.types import VectorStoreQuery, VectorStoreQueryMode

# Assume query_embedding is the embedding you obtained
query = VectorStoreQuery(
    query_embedding=query_embedding,
    similarity_top_k=10,  # top 10 most similar nodes
    mode=VectorStoreQueryMode.DEFAULT
)

result = MilvusVectorStore().query(query)

# result.nodes will contain the most similar nodes
# result.similarities will contain the similarities
# result.ids will contain the ids

Please note that you need to replace MilvusVectorStore() with an instance of your specific vector store class that inherits from VectorStore.

The VectorStoreQuery object is a dataclass that represents a query to the vector store. It has the following structure and parameters:

• query_embedding: An optional list of floats. This represents the embedding of the query.
• similarity_top_k: An integer. This represents the number of top similar results to return.
• doc_ids: An optional list of strings. These are the document IDs to query.
• node_ids: An optional list of strings. These are the node IDs to query.
• query_str: An optional string. This is the query string.
• output_fields: An optional list of strings. These are the fields to output in the result.
• embedding_field: An optional string. This is the field that contains the embedding.
• mode: A VectorStoreQueryMode enum. This represents the mode of the query. The default is VectorStoreQueryMode.DEFAULT.
• alpha: An optional float. This is only for hybrid search (0 for bm25, 1 for vector search).
• filters: An optional MetadataFilters object. These are the metadata filters to apply to the query.
• mmr_threshold: An optional float. This is only for maximum marginal relevance (mmr) mode.
• sparse_top_k: An optional integer. This is currently only used by postgres hybrid search.

You can find more details in the source code:

• llama_index/embeddings/base.py
• llama_index/vector_stores/milvus.py
• llama_index/vector_stores/types.py

I hope this helps! If you have any other questions, feel free to ask.

Best,
Dosu

Sources

• llama_index/embeddings/base.py
• llama_index/vector_stores/milvus.py
• llama_index/vector_stores/types.py

---

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.

[axz91]

🤖

Hello,

To query based on a particular embedding in the LlamaIndex database after obtaining it from a collection using LLM, you can use the query method provided in the MilvusVectorStore class. This method takes a VectorStoreQuery object as input and returns a VectorStoreQueryResult object which contains the most similar nodes, their similarities, and their ids.

Here is an example of how you can use it:

from llama_index.vector_stores.types import VectorStoreQuery, VectorStoreQueryMode

# Assume query_embedding is the embedding you obtained
query = VectorStoreQuery(
    query_embedding=query_embedding,
    similarity_top_k=10,  # top 10 most similar nodes
    mode=VectorStoreQueryMode.DEFAULT
)

result = MilvusVectorStore().query(query)

# result.nodes will contain the most similar nodes
# result.similarities will contain the similarities
# result.ids will contain the ids

Please note that you need to replace MilvusVectorStore() with an instance of your specific vector store class that inherits from VectorStore.

The VectorStoreQuery object is a dataclass that represents a query to the vector store. It has the following structure and parameters:

• query_embedding: An optional list of floats. This represents the embedding of the query.
• similarity_top_k: An integer. This represents the number of top similar results to return.
• doc_ids: An optional list of strings. These are the document IDs to query.
• node_ids: An optional list of strings. These are the node IDs to query.
• query_str: An optional string. This is the query string.
• output_fields: An optional list of strings. These are the fields to output in the result.
• embedding_field: An optional string. This is the field that contains the embedding.
• mode: A VectorStoreQueryMode enum. This represents the mode of the query. The default is VectorStoreQueryMode.DEFAULT.
• alpha: An optional float. This is only for hybrid search (0 for bm25, 1 for vector search).
• filters: An optional MetadataFilters object. These are the metadata filters to apply to the query.
• mmr_threshold: An optional float. This is only for maximum marginal relevance (mmr) mode.
• sparse_top_k: An optional integer. This is currently only used by postgres hybrid search.

You can find more details in the source code:

• llama_index/embeddings/base.py
• llama_index/vector_stores/milvus.py
• llama_index/vector_stores/types.py

I hope this helps! If you have any other questions, feel free to ask.

Best, Dosu

Sources

• llama_index/embeddings/base.py
• llama_index/vector_stores/milvus.py
• llama_index/vector_stores/types.py

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.

how to define what LLM is being used

[dosubot[bot]]

🤖

Hello,

Yes, you are on the right track. Your understanding of the query method in the MilvusVectorStore class and the VectorStoreQuery object is correct.

The query method takes a VectorStoreQuery object as an argument, which contains the query embedding, the number of most similar nodes to return (similarity_top_k), and optional filters for specific document IDs (doc_ids) and node IDs (node_ids).

Here is the relevant code:

def query(self, query: VectorStoreQuery, **kwargs: Any) -> VectorStoreQueryResult:
    """Query index for top k most similar nodes.

    Args:
        query_embedding (List[float]): query embedding
        similarity_top_k (int): top k most similar nodes
        doc_ids (Optional[List[str]]): list of doc_ids to filter by
        node_ids (Optional[List[str]]): list of node_ids to filter by
        output_fields (Optional[List[str]]): list of fields to return
        embedding_field (Optional[str]): name of embedding field
    """
    if query.mode != VectorStoreQueryMode.DEFAULT:
        raise ValueError(f"Milvus does not support {query.mode} yet.")

    expr = []
    output_fields = ["*"]

    # Parse the filter
    if query.filters is not None:
        expr.extend(_to_milvus_filter(query.filters))

    # Parse any docs we are filtering on
    if query.doc_ids is not None and len(query.doc_ids) != 0:
        expr_list = ['"' + entry + '"' for entry in query.doc_ids]
        expr.append(f"{self.doc_id_field} in [{','.join(expr_list)}]")
    # Parse any nodes we are filtering on
    if query.node_ids is not None and len(query.node_ids) != 0:
        expr_list = ['"' + entry + '"' for entry in query.node_ids]
        expr.append(f"{MILVUS_ID_FIELD} in [{','.join(expr_list)}]")

    # Limit output fields
    if query.output_fields is not None:
        output_fields = query.output_fields

    # Convert to string expression
    string_expr = ""
    if len(expr) != 0:
        string_expr = " and ".join(expr)

    # Perfom the search
    res = self.milvusclient.search(
        collection_name=self.collection_name,
        data=[query.query_embedding],
        filter=string_expr,
        limit=query.similarity_top_k,
        output_fields=output_fields,
    )

    logger.debug(
        f"Successfully searched embedding in collection: {self.collection_name}"
        f" Num Results: {len(res[0])}"
    )

    nodes = []
    similarities = []
    ids = []

    # Parse the results
    for hit in res[0]:
        node = metadata_dict_to_node(
            {"_node_content": hit["entity"].get("_node_content", None)}
        )
        nodes.append(node)
        similarities.append(hit["distance"])
        ids.append(hit["id"])

    return VectorStoreQueryResult(nodes=nodes, similarities=similarities, ids=ids)

The query method first checks if the query mode is supported by Milvus. It then constructs a filter expression based on the provided filters, document IDs, and node IDs. The method then performs the search on the Milvus collection using the query embedding and the constructed filter expression. The results are parsed into a VectorStoreQueryResult object, which contains the most similar nodes, their similarities, and their IDs.

There are no additional steps you need to take unless you want to filter the results by specific document IDs or node IDs, in which case you would include these in the VectorStoreQuery object.

As for the VectorStoreQueryMode, it is used to specify the mode of querying in the Vector Store. The possible values are DEFAULT, SPARSE, HYBRID, TEXT_SEARCH, SVM, LOGISTIC_REGRESSION, LINEAR_REGRESSION, and MMR. However, please note that the query method in the MilvusVectorStore class currently only supports the DEFAULT mode.

I hope this helps! If you have any other questions, feel free to ask.

Best,
Dosu

Sources

• llama_index/vector_stores/milvus.py
• llama_index/vector_stores/types.py

---

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.