File size: 3,085 Bytes
f9aa233
 
 
 
 
 
 
 
566cab6
f9aa233
 
 
 
 
67b6f0a
 
 
 
f9aa233
 
 
 
 
 
 
 
 
 
 
 
21e2f95
f9aa233
 
 
 
 
21e2f95
f9aa233
21e2f95
f9aa233
 
21e2f95
f9aa233
21e2f95
 
 
 
 
 
 
 
04af09b
21e2f95
 
 
 
 
 
 
 
f9aa233
 
 
 
67b6f0a
f9aa233
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
---
pretty_name: Example Documents
tags:
- sentence-transformers
---

# Example Documents

A small set of example documents across modalities (image, audio, video) for use in [Sentence Transformers](https://www.sbert.net/) retrieval snippets and documentation. These are the kinds of files you pass to `model.encode_document(...)`. They can safely be used as examples in your model cards if you don't want to host the example assets in your model repositories themselves.

## Contents

| File | Modality |
| --- | --- |
| `doc1.jpg` | image (document page) |
| `doc2.jpg` | image (document page) |
| `doc3.jpg` | image (document page) |
| `doc4.jpg` | image (document page) |
| `llama4_hgf.png` | image |
| `qwen2.5omni_hgf.png` | image |
| `jay_chou_superman_cant_fly.mp3` | audio (music) |
| `joe_hisaishi_summer.mp3` | audio (music) |
| `conversation1.mp3` | audio (speech) |
| `conversation2.mp3` | audio (speech) |
| `conversation3.mp3` | audio (speech) |
| `mapo_tofu.mp4` | video |
| `zhajiang_noodle.mp4` | video |

## Usage

Reference any file by its resolve URL. These documents can be encoded with a multi-vector (late interaction) `MultiVectorEncoder`:

```python
from sentence_transformers import MultiVectorEncoder

model = MultiVectorEncoder("vidore/colqwen-omni-v0.1")
queries = ["What is the Llama 4 model?"]
documents = [
    "https://huggingface.co/datasets/sentence-transformers/example-documents/resolve/main/llama4_hgf.png",
    "https://huggingface.co/datasets/sentence-transformers/example-documents/resolve/main/conversation3.mp3",
]
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
print(model.similarity(query_embeddings, document_embeddings))
```

or with a single-vector `SentenceTransformer`:

```python
from sentence_transformers import SentenceTransformer

model = SentenceTransformer("LCO-Embedding/LCO-Embedding-Omni-3B-2605")
queries = ["What is the Llama 4 model?"]
documents = [
    "https://huggingface.co/datasets/sentence-transformers/example-documents/resolve/main/llama4_hgf.png",
    "https://huggingface.co/datasets/sentence-transformers/example-documents/resolve/main/conversation3.mp3",
]
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
print(model.similarity(query_embeddings, document_embeddings))
```

## Credits

- The document page images (`doc1.jpg` to `doc4.jpg`) are the first four test documents from [vidore/colpali_train_set](https://huggingface.co/datasets/vidore/colpali_train_set).
- The images (`llama4_hgf.png`, `qwen2.5omni_hgf.png`), music (`jay_chou_superman_cant_fly.mp3`, `joe_hisaishi_summer.mp3`), and videos (`mapo_tofu.mp4`, `zhajiang_noodle.mp4`) are copied from [Tevatron/OmniEmbed-v0.1](https://huggingface.co/Tevatron/OmniEmbed-v0.1). Thanks to the Tevatron team.
- The speech clips (`conversation1.mp3`, `conversation2.mp3`, `conversation3.mp3`) are short (about 30 second) excerpts from [eustlb/dailytalk-conversations-grouped](https://huggingface.co/datasets/eustlb/dailytalk-conversations-grouped).