djovak commited on
Commit
d1a532e
1 Parent(s): dd1b86e

update README.md

Browse files
Files changed (3) hide show
  1. README.md +50 -7
  2. image-1.png +0 -0
  3. image-2.png +0 -0
README.md CHANGED
@@ -2,14 +2,25 @@
2
  library_name: sentence-transformers
3
  pipeline_tag: sentence-similarity
4
  tags:
 
5
  - sentence-transformers
6
  - feature-extraction
7
  - sentence-similarity
8
-
 
 
 
 
9
  ---
10
 
11
  # djovak/embedic-small
12
 
 
 
 
 
 
 
13
  This is a [sentence-transformers](https://www.SBERT.net) model: It maps sentences & paragraphs to a 384 dimensional dense vector space and can be used for tasks like clustering or semantic search.
14
 
15
  <!--- Describe your model here -->
@@ -26,22 +37,54 @@ Then you can use the model like this:
26
 
27
  ```python
28
  from sentence_transformers import SentenceTransformer
29
- sentences = ["This is an example sentence", "Each sentence is converted"]
30
 
31
  model = SentenceTransformer('djovak/embedic-small')
32
  embeddings = model.encode(sentences)
33
  print(embeddings)
34
  ```
35
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
36
 
 
37
 
38
- ## Evaluation Results
39
 
40
- <!--- Describe how your model was evaluated -->
41
 
42
- For an automated evaluation of this model, see the *Sentence Embeddings Benchmark*: [https://seb.sbert.net](https://seb.sbert.net?model_name=djovak/embedic-small)
43
 
 
44
 
 
45
 
46
  ## Full Model Architecture
47
  ```
@@ -52,6 +95,6 @@ SentenceTransformer(
52
  )
53
  ```
54
 
55
- ## Citing & Authors
56
 
57
- <!--- Describe where people can find more information -->
 
2
  library_name: sentence-transformers
3
  pipeline_tag: sentence-similarity
4
  tags:
5
+ - mteb
6
  - sentence-transformers
7
  - feature-extraction
8
  - sentence-similarity
9
+ license: mit
10
+ language:
11
+ - multilingual
12
+ - en
13
+ - sr
14
  ---
15
 
16
  # djovak/embedic-small
17
 
18
+ Say hello to **Embedić**, a group of new text embedding models finetuned for the Serbian language!
19
+
20
+ These models are particularly useful in Information Retrieval and RAG purposes. Check out images showcasing benchmark performance, you can beat previous SOTA with 5x fewer parameters!
21
+
22
+ Although specialized for Serbian(Cyrillic and Latin scripts), Embedić is Cross-lingual(it understands English too). So you can embed English docs, Serbian docs, or a combination of the two :)
23
+
24
  This is a [sentence-transformers](https://www.SBERT.net) model: It maps sentences & paragraphs to a 384 dimensional dense vector space and can be used for tasks like clustering or semantic search.
25
 
26
  <!--- Describe your model here -->
 
37
 
38
  ```python
39
  from sentence_transformers import SentenceTransformer
40
+ sentences = ["ko je Nikola Tesla?", "Nikola Tesla je poznati pronalazač", "Nikola Jokić je poznati košarkaš"]
41
 
42
  model = SentenceTransformer('djovak/embedic-small')
43
  embeddings = model.encode(sentences)
44
  print(embeddings)
45
  ```
46
 
47
+ ### Important usage notes
48
+ - "ošišana ćirilica" (usage of c instead of ć, etc...) significantly deacreases search quality
49
+ - The usage of uppercase letters for named entities can significantly improve search quality
50
+
51
+
52
+ ## Evaluation
53
+
54
+
55
+ ### **Model description**:
56
+
57
+ | Model Name | Dimension | Sequence Length | Parameters
58
+ |:----:|:---:|:---:|:---:|
59
+ | [intfloat/multilingual-e5-small](https://huggingface.co/intfloat/multilingual-e5-small) | 384 | 512 | 117M
60
+ | [djovak/embedic-small](https://huggingface.co/djovak/embedic-small) | 384 | 512 | 117M
61
+ |||||||||
62
+ | [intfloat/multilingual-e5-base](https://huggingface.co/intfloat/multilingual-e5-base) | 768 | 512 | 278M
63
+ | [djovak/embedic-base](https://huggingface.co/djovak/embedic-base) | 768 | 512 | 278M
64
+ |||||||||
65
+ | [intfloat/multilingual-e5-large](https://huggingface.co/intfloat/multilingual-e5-large) | 1024 | 512 | 560M
66
+ | [djovak/embedic-large](https://huggingface.co/djovak/embedic-large) | 1024 | 512 | 560M
67
+
68
+
69
+
70
+ `BM25-ENG` - Elasticsearch with English analyzer
71
+
72
+
73
+ `BM25-SRB` - Elasticsearch with Serbian analyzer
74
+
75
+ ### evaluation resultsresults
76
 
77
+ Evaluation on 3 tasks: Information Retrieval, Sentence Similarity, and Bitext mining. I personally translated the STS17 cross-lingual evaluation dataset and Spent 6,000$ on Google translate API, translating 4 IR evaluation datasets into Serbian language.
78
 
79
+ Evaluation datasets will be published as Part of [MTEB benchmark](https://huggingface.co/spaces/mteb/leaderboard) in the near future.
80
 
81
+ ![information retrieval results](image-2.png)
82
 
83
+ ![sentence similarity results](image-1.png)
84
 
85
+ ## Contact
86
 
87
+ If you have any question or sugestion related to this project, you can open an issue or pull request. You can also email me at novakzivanic@gmail.com
88
 
89
  ## Full Model Architecture
90
  ```
 
95
  )
96
  ```
97
 
98
+ ## License
99
 
100
+ Embedić models are licensed under the [MIT License](https://github.com/FlagOpen/FlagEmbedding/blob/master/LICENSE). The released models can be used for commercial purposes free of charge.
image-1.png ADDED
image-2.png ADDED