daft-qdrant 0.2.0

Qdrant community extension for Daft

daft-qdrant

Community extension for Daft to write vector embeddings and their payloads into Qdrant collections.

Installation

pip install daft-qdrant

Quick start

import daft
import daft_qdrant  # registers DataFrame.write_qdrant

df = daft.from_pydict({
    "id": [1, 2, 3],
    "vector": [[0.1, 0.2, 0.3], [0.4, 0.5, 0.6], [0.7, 0.8, 0.9]],
    "label": ["cat", "dog", "bird"],
})

df.write_qdrant("my-collection", url="http://localhost:6333")

The target collection must exist before writing. Each row becomes a Qdrant point:

• id column → point id (unsigned integer or UUID string)
• vector column → point vector (list of floats, or dict for named vectors)
• all other columns → point payload

Embedding pipeline example

import os
import daft
import daft_qdrant
from daft.functions import monotonically_increasing_id
from daft.functions.ai import embed_text
from qdrant_client import QdrantClient, models

url = "http://localhost:6333"
vector_size = 768

QdrantClient(url=url).create_collection(
    "daft-qdrant-example",
    vectors_config=models.VectorParams(size=vector_size, distance=models.Distance.COSINE),
)

(
    daft.read_huggingface("Open-Orca/OpenOrca")
    .limit(100)
    .with_column(
        "vector",
        embed_text(daft.col("response"), provider="sentence_transformers", model="BAAI/bge-base-en-v1.5"),
    )
    .with_column("id", monotonically_increasing_id())
    .write_qdrant("daft-qdrant-example", url=url)
)

Routing rows to multiple collections

Pass a Daft expression as collection_name to route rows to different collections based on a column value:

df.write_qdrant(daft.col("collection"), url="http://localhost:6333")

Column remapping

Use id_column and vector_column to map differently named columns to the required id and vector roles:

df.write_qdrant(
    "my-collection",
    url="http://localhost:6333",
    id_column="doc_id",
    vector_column="embedding",
)

Links

• Daft documentation
• Community extensions
• Qdrant documentation