chadhas commited on
Commit
0cd7d65
·
verified ·
1 Parent(s): 1eac865

Upload folder using huggingface_hub

Browse files
Files changed (2) hide show
  1. README.md +121 -0
  2. data/train-00000-of-00001.parquet +3 -0
README.md ADDED
@@ -0,0 +1,121 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-4.0
3
+ language:
4
+ - en
5
+ pretty_name: Qdrant documentation with multilingual-e5-small embeddings
6
+ task_categories:
7
+ - text-retrieval
8
+ - feature-extraction
9
+ size_categories:
10
+ - 1K<n<10K
11
+ tags:
12
+ - qdrant
13
+ - hybrid-search
14
+ ---
15
+
16
+ # qdrant-docs-multilingual-e5-small-embeddings
17
+
18
+ This dataset contains the Qdrant documentation, courses, and articles from [qdrant.tech](https://qdrant.tech),
19
+ as of October 25, 2026, split into 3,593 chunks by page section. A section that fits in 480 tokens becomes one
20
+ chunk, and larger sections are split further. Code examples are kept in Python only.
21
+
22
+ Each row holds a chunk's ID, its payload, and a 384-dimensional embedding:
23
+
24
+ | Column | Type | Description |
25
+ |---|---|---|
26
+ | `id` | string | Point ID in the Qdrant collection |
27
+ | `url` | string | Page URL, relative to qdrant.tech |
28
+ | `title` | string | Page title |
29
+ | `section` | string | Headings above the chunk, joined with " > ". Empty for page intros |
30
+ | `content_type` | string | `documentation`, `course`, or `article` |
31
+ | `has_code` | bool | Whether the chunk contains a code block |
32
+ | `text` | string | A heading line, then the chunk in Markdown |
33
+ | `vector` | list of float32 | multilingual-e5-small embedding, 384 dimensions, cosine |
34
+
35
+ ## Generation process
36
+
37
+ The embeddings were generated with the [multilingual-e5-small](https://huggingface.co/intfloat/multilingual-e5-small)
38
+ model from the `text` field with fenced code blocks, Markdown images, and link URLs removed, using the `passage: `
39
+ prefix. Embed your search queries with the `query: ` prefix instead:
40
+
41
+ ```python
42
+ from sentence_transformers import SentenceTransformer
43
+
44
+ model = SentenceTransformer("intfloat/multilingual-e5-small")
45
+ text = "Final Project: Production-Ready Documentation Search Engine > Build Steps > Step 3: Parse and Chunk Documents"
46
+ embeddings = model.encode([f"passage: {text}"], normalize_embeddings=True)
47
+ query_embedding = model.encode(
48
+ ["how to chunk documentation"], prompt="query: ", normalize_embeddings=True
49
+ )
50
+ ```
51
+
52
+ ## Using the dataset with Qdrant
53
+
54
+ The quickest way is to restore the ready-made [snapshot](https://snapshots.qdrant.io/qdrant-docs-2026-10-25.snapshot),
55
+ which includes the payload indexes. It requires Qdrant 1.20 or later:
56
+
57
+ ```http
58
+ PUT /collections/qdrant-docs/snapshots/recover
59
+ {
60
+ "location": "https://snapshots.qdrant.io/qdrant-docs-2026-10-25.snapshot"
61
+ }
62
+ ```
63
+
64
+ To load the rows yourself, create the collection with a vector named `dense`, upload the rows, then create the
65
+ payload indexes. The full-text index on `text` uses BM25 scoring, which also requires Qdrant 1.20 or later:
66
+
67
+ ```python
68
+ from datasets import load_dataset
69
+ from qdrant_client import QdrantClient, models
70
+
71
+ client = QdrantClient(url="http://localhost:6333")
72
+ client.create_collection(
73
+ "qdrant-docs",
74
+ vectors_config={
75
+ "dense": models.VectorParams(size=384, distance=models.Distance.COSINE)
76
+ },
77
+ # The collection is small, so lower the threshold to build the HNSW graph anyway.
78
+ optimizers_config=models.OptimizersConfigDiff(indexing_threshold=1000),
79
+ )
80
+
81
+ dataset = load_dataset(
82
+ "Qdrant/qdrant-docs-multilingual-e5-small-embeddings", split="train"
83
+ )
84
+ for batch in dataset.iter(batch_size=256):
85
+ payload_fields = [name for name in batch if name not in ("id", "vector")]
86
+ client.upsert(
87
+ "qdrant-docs",
88
+ points=[
89
+ models.PointStruct(
90
+ id=batch["id"][i],
91
+ vector={"dense": batch["vector"][i]},
92
+ payload={name: batch[name][i] for name in payload_fields},
93
+ )
94
+ for i in range(len(batch["id"]))
95
+ ],
96
+ )
97
+
98
+ for field in ("url", "content_type"):
99
+ client.create_payload_index("qdrant-docs", field, models.PayloadSchemaType.KEYWORD)
100
+ client.create_payload_index("qdrant-docs", "has_code", models.PayloadSchemaType.BOOL)
101
+ client.create_payload_index(
102
+ "qdrant-docs",
103
+ "text",
104
+ models.TextIndexParams(
105
+ type=models.TextIndexType.TEXT,
106
+ tokenizer=models.TokenizerType.CODE,
107
+ lowercase=True,
108
+ stemmer=models.SnowballParams(
109
+ type=models.Snowball.SNOWBALL, language=models.SnowballLanguage.ENGLISH
110
+ ),
111
+ stopwords=models.Language.ENGLISH,
112
+ scoring=models.TextScoringParams(type=models.TextScoringType.BM25),
113
+ ),
114
+ )
115
+ ```
116
+
117
+ Pass `using="dense"` when you query the vector.
118
+
119
+ ## License
120
+
121
+ The content belongs to Qdrant and is released under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/).
data/train-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d95e62c65728ffee93ea3cad3e8126356b12a187525b7e31e10abcfaa066ed74
3
+ size 10557880