README.md
5.0 KB · 184 lines · markdown Raw
1 ---
2 language:
3 - multilingual
4 - ar
5 - bg
6 - ca
7 - cs
8 - da
9 - de
10 - el
11 - en
12 - es
13 - et
14 - fa
15 - fi
16 - fr
17 - gl
18 - gu
19 - he
20 - hi
21 - hr
22 - hu
23 - hy
24 - id
25 - it
26 - ja
27 - ka
28 - ko
29 - ku
30 - lt
31 - lv
32 - mk
33 - mn
34 - mr
35 - ms
36 - my
37 - nb
38 - nl
39 - pl
40 - pt
41 - ro
42 - ru
43 - sk
44 - sl
45 - sq
46 - sr
47 - sv
48 - th
49 - tr
50 - uk
51 - ur
52 - vi
53 license: apache-2.0
54 library_name: sentence-transformers
55 tags:
56 - sentence-transformers
57 - feature-extraction
58 - sentence-similarity
59 - transformers
60 - text-embeddings-inference
61 language_bcp47:
62 - fr-ca
63 - pt-br
64 - zh-cn
65 - zh-tw
66 pipeline_tag: sentence-similarity
67 ---
68
69 # sentence-transformers/paraphrase-multilingual-mpnet-base-v2
70
71 This is a [sentence-transformers](https://www.SBERT.net) model: It maps sentences & paragraphs to a 768 dimensional dense vector space and can be used for tasks like clustering or semantic search.
72
73
74
75 ## Usage (Sentence-Transformers)
76
77 Using this model becomes easy when you have [sentence-transformers](https://www.SBERT.net) installed:
78
79 ```
80 pip install -U sentence-transformers
81 ```
82
83 Then you can use the model like this:
84
85 ```python
86 from sentence_transformers import SentenceTransformer
87 sentences = ["This is an example sentence", "Each sentence is converted"]
88
89 model = SentenceTransformer('sentence-transformers/paraphrase-multilingual-mpnet-base-v2')
90 embeddings = model.encode(sentences)
91 print(embeddings)
92 ```
93
94
95
96 ## Usage (HuggingFace Transformers)
97 Without [sentence-transformers](https://www.SBERT.net), you can use the model like this: First, you pass your input through the transformer model, then you have to apply the right pooling-operation on-top of the contextualized word embeddings.
98
99 ```python
100 from transformers import AutoTokenizer, AutoModel
101 import torch
102
103
104 # Mean Pooling - Take attention mask into account for correct averaging
105 def mean_pooling(model_output, attention_mask):
106 token_embeddings = model_output[0] # First element of model_output contains all token embeddings
107 input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
108 return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)
109
110
111 # Sentences we want sentence embeddings for
112 sentences = ['This is an example sentence', 'Each sentence is converted']
113
114 # Load model from HuggingFace Hub
115 tokenizer = AutoTokenizer.from_pretrained('sentence-transformers/paraphrase-multilingual-mpnet-base-v2')
116 model = AutoModel.from_pretrained('sentence-transformers/paraphrase-multilingual-mpnet-base-v2')
117
118 # Tokenize sentences
119 encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')
120
121 # Compute token embeddings
122 with torch.no_grad():
123 model_output = model(**encoded_input)
124
125 # Perform pooling. In this case, mean pooling
126 sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])
127
128 print("Sentence embeddings:")
129 print(sentence_embeddings)
130 ```
131
132
133 ## Usage (Text Embeddings Inference (TEI))
134
135 [Text Embeddings Inference (TEI)](https://github.com/huggingface/text-embeddings-inference) is a blazing fast inference solution for text embedding models.
136
137 - CPU:
138 ```bash
139 docker run -p 8080:80 -v hf_cache:/data --pull always ghcr.io/huggingface/text-embeddings-inference:cpu-latest --model-id sentence-transformers/paraphrase-multilingual-mpnet-base-v2 --pooling mean --dtype float16
140 ```
141
142 - NVIDIA GPU:
143 ```bash
144 docker run --gpus all -p 8080:80 -v hf_cache:/data --pull always ghcr.io/huggingface/text-embeddings-inference:cuda-latest --model-id sentence-transformers/paraphrase-multilingual-mpnet-base-v2 --pooling mean --dtype float16
145 ```
146
147 Send a request to `/v1/embeddings` to generate embeddings via the [OpenAI Embeddings API](https://platform.openai.com/docs/api-reference/embeddings/create):
148 ```bash
149 curl http://localhost:8080/v1/embeddings \
150 -H "Content-Type: application/json" \
151 -d '{
152 "model": "sentence-transformers/paraphrase-multilingual-mpnet-base-v2",
153 "input": "This is an example sentence"
154 }'
155 ```
156
157 Or check the [Text Embeddings Inference API specification](https://huggingface.github.io/text-embeddings-inference/) instead.
158
159
160
161 ## Full Model Architecture
162 ```
163 SentenceTransformer(
164 (0): Transformer({'max_seq_length': 128, 'do_lower_case': False}) with Transformer model: XLMRobertaModel
165 (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False})
166 )
167 ```
168
169 ## Citing & Authors
170
171 This model was trained by [sentence-transformers](https://www.sbert.net/).
172
173 If you find this model helpful, feel free to cite our publication [Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks](https://arxiv.org/abs/1908.10084):
174 ```bibtex
175 @inproceedings{reimers-2019-sentence-bert,
176 title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
177 author = "Reimers, Nils and Gurevych, Iryna",
178 booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
179 month = "11",
180 year = "2019",
181 publisher = "Association for Computational Linguistics",
182 url = "http://arxiv.org/abs/1908.10084",
183 }
184 ```