asi commited on
Commit
ba54e99
1 Parent(s): 9067b00

:books: add documentation

Browse files
Files changed (1) hide show
  1. README.md +89 -7
README.md CHANGED
@@ -1,11 +1,93 @@
1
  ---
2
- pipeline_tag: sentence-similarity
3
- tags:
4
- - sentence-transformers
5
- - feature-extraction
6
- - sentence-similarity
7
  ---
8
 
9
- # all_datasets_v3_MiniLM-L12
10
 
11
- This is a MiniLM-L12-H384 model trained on all the dataset of the 1B+ train corpus. It was trained with the v3 setup. See data_config.json and train_script.py in this respository how the model was trained and which datasets have been used.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ language: en
 
 
 
 
3
  ---
4
 
 
5
 
6
+ # Model description
7
+
8
+ The project aims to train sentence embedding models on very large sentence level datasets using a self-supervised
9
+ contrastive learning objective. We used the pretrained [`MiniLM-L12`](https://huggingface.co/microsoft/MiniLM-L12-H384-uncased) model and fine-tuned in on a
10
+ 1B sentence pairs dataset. We use a contrastive learning objective: given a sentence from the pair, the model should predict which out of a set of randomly sampled other sentences, was actually paired with it in our dataset.
11
+
12
+ We developped this model during the
13
+ [Community week using JAX/Flax for NLP & CV](https://discuss.huggingface.co/t/open-to-the-community-community-week-using-jax-flax-for-nlp-cv/7104),
14
+ organized by Hugging Face. We developped this model as part of the project:
15
+ [Train the Best Sentence Embedding Model Ever with 1B Training Pairs](https://discuss.huggingface.co/t/train-the-best-sentence-embedding-model-ever-with-1b-training-pairs/7354). We benefited from efficient hardware infrastructure to run the project: 7 TPUs v3-8, as well
16
+ as intervention from Google’s Flax, JAX, and Cloud team member about efficient deep learning frameworks.
17
+
18
+ ## Intended uses
19
+
20
+ Our model is intented to be used as a sentence encoder. Given an input sentence, it ouptuts a vector which captures
21
+ the sentence semantic information. The sentence vector may be used for information retrieval, clustering or sentence
22
+ similarity tasks.
23
+
24
+ ## How to use
25
+
26
+ Here is how to use this model to get the features of a given text using [SentenceTransformers](https://github.com/UKPLab/sentence-transformers) library:
27
+
28
+ ```python
29
+ from sentence_transformers import SentenceTransformer
30
+
31
+ model = SentenceTransformer('flax-sentence-embeddings/all_datasets_v3_MiniLM-L12')
32
+ text = "Replace me by any text you'd like."
33
+ text_embbedding = model.encode(text)
34
+ # array([-0.01559514, 0.04046123, 0.1317083 , 0.00085931, 0.04585106,
35
+ # -0.05607086, 0.0138078 , 0.03569756, 0.01420381, 0.04266302 ...],
36
+ # dtype=float32)
37
+ ```
38
+
39
+ # Training procedure
40
+
41
+ ## Pre-training
42
+
43
+ We use the pretrained [`MiniLM-L12`](https://huggingface.co/microsoft/MiniLM-L12-H384-uncased). Please refer to the model
44
+ card for more detailed information about the pre-training procedure.
45
+
46
+ ## Fine-tuning
47
+
48
+ The BERT model was pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
49
+ unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
50
+ headers).
51
+
52
+ ### Hyper parameters
53
+
54
+ We trained ou model on a TPU v3-8. We train the model during 540k steps using a batch size of 1024 (128 per TPU core).
55
+ We use a learning rate warm up of 500. The sequence length was limited to 128 tokens. We used the AdamW optimizer with
56
+ a 2e-5 learning rate. The full training script is accessible in this current repository.
57
+
58
+ ### Training data
59
+
60
+ We use the concatenation from multiple datasets to fine-tune our model. The total number of sentence pairs is above 1 billion sentences.
61
+ We sampled each dataset given a weighted probability which configuration is detailed in the `data_config.json` file.
62
+
63
+
64
+ | Dataset | Paper | Number of training tuples |
65
+ |:--------------------------------------------------------:|:----------------------------------------:|:--------------------------:|
66
+ | [GOOAQ: Open Question Answering with Diverse Answer Types](https://github.com/allenai/gooaq) | https://arxiv.org/pdf/2104.08727.pdf | 3,012,496 |
67
+ | [Stack Exchange](https://huggingface.co/datasets/flax-sentence-embeddings/stackexchange_title_body_jsonl) | - | 364,001 |
68
+ | [Flickr 30k](https://shannon.cs.illinois.edu/DenotationGraph/) | https://transacl.org/ojs/index.php/tacl/article/view/229/33 | 317,695 |
69
+ | [COCO 2020](COCO 2020) | https://link.springer.com/chapter/10.1007%2F978-3-319-10602-1_48 | 828,395|
70
+ | [Code Search](https://huggingface.co/datasets/code_search_net) | - | 1,151,414 |
71
+ | [TriviaqQA](https://huggingface.co/datasets/trivia_qa) | - | 73,346 |
72
+ | [SQuAD2.0](https://rajpurkar.github.io/SQuAD-explorer/) | https://aclanthology.org/P18-2124.pdf | 87,599 |
73
+ | [Natural Questions (NQ)](https://ai.google.com/research/NaturalQuestions) | https://transacl.org/ojs/index.php/tacl/article/view/1455 | 100,231 |
74
+ | [Simple Wikipedia](https://cs.pomona.edu/~dkauchak/simplification/) | https://www.aclweb.org/anthology/P11-2117/ | 102,225 |
75
+ | [Quora Question Pairs](https://quoradata.quora.com/First-Quora-Dataset-Release-Question-Pairs) | - | 103,663 |
76
+ | [Altlex](https://github.com/chridey/altlex/) | https://aclanthology.org/P16-1135.pdf | 112,696 |
77
+ | [Wikihow](https://github.com/pvl/wikihow_pairs_dataset) | https://arxiv.org/abs/1810.09305 | 128,542 |
78
+ | [Sentence Compression](https://github.com/google-research-datasets/sentence-compression) | https://www.aclweb.org/anthology/D13-1155/ | 180,000 |
79
+ | AllNLI ([SNLI](https://nlp.stanford.edu/projects/snli/) and [MultiNLI](https://cims.nyu.edu/~sbowman/multinli/) | https://doi.org/10.18653/v1/d15-1075, https://doi.org/10.18653/v1/n18-1101 | 277,230 |
80
+ | [Eli5](https://huggingface.co/datasets/eli5) | https://doi.org/10.18653/v1/p19-1346 | 325,475 |
81
+ | [SPECTER](https://github.com/allenai/specter) | https://doi.org/10.18653/v1/2020.acl-main.207 | 684,100 |
82
+ | [S2ORC](https://github.com/allenai/s2orc) Title/Abstract | https://aclanthology.org/2020.acl-main.447/ | 41,769,185 |
83
+ | [S2ORC](https://github.com/allenai/s2orc) Citation/Citation | https://aclanthology.org/2020.acl-main.447/ | 52,603,982 |
84
+ | [S2ORC](https://github.com/allenai/s2orc) Citation/Abstract | https://aclanthology.org/2020.acl-main.447/ | 116,288,806 |
85
+ | [PAQ](https://github.com/facebookresearch/PAQ) | https://arxiv.org/abs/2102.07033 | 64,371,441 |
86
+ | [WikiAnswers](https://github.com/afader/oqa#wikianswers-corpus) | https://doi.org/10.1145/2623330.2623677 | 77,427,422 |
87
+ | SearchQA | - | 582,261 |
88
+ | [Yahoo Answers](https://www.kaggle.com/soumikrakshit/yahoo-answers-dataset) Title/Answer | https://proceedings.neurips.cc/paper/2015/hash/250cf8b51c773f3f8dc8b4be867a9a02-Abstract.html | 1,198,260 |
89
+ | [Yahoo Answers](https://www.kaggle.com/soumikrakshit/yahoo-answers-dataset) Title/Question | https://proceedings.neurips.cc/paper/2015/hash/250cf8b51c773f3f8dc8b4be867a9a02-Abstract.html | 659,896 |
90
+ | [Yahoo Answers](https://www.kaggle.com/soumikrakshit/yahoo-answers-dataset) Question/Answer | https://proceedings.neurips.cc/paper/2015/hash/250cf8b51c773f3f8dc8b4be867a9a02-Abstract.html | 681,164 |
91
+ | [MS MARCO](https://microsoft.github.io/msmarco/) | https://doi.org/10.1145/3404835.3462804 | 9,144,553 |
92
+ | [Reddit conversationnal](https://github.com/PolyAI-LDN/conversational-datasets/tree/master/reddit) | https://arxiv.org/abs/1904.06472 | 726,484,430 |
93
+