From 52d250f6aa14844024806e5e4dd1c7882bbd8dd5 Mon Sep 17 00:00:00 2001
From: Pedro Lima <pedro.lima@gmail.com>
Date: Tue, 15 Sep 2020 13:54:12 +0100
Subject: [PATCH] [model_cards] pvl/labse_bert model card

From **Language-Agnostic BERT Sentence Embedding**

https://ai.googleblog.com/2020/08/language-agnostic-bert-sentence.html
---
 model_cards/pvl/labse_bert/README.md | 47 ++++++++++++++++++++++++++++
 1 file changed, 47 insertions(+)
 create mode 100644 model_cards/pvl/labse_bert/README.md

diff --git a/model_cards/pvl/labse_bert/README.md b/model_cards/pvl/labse_bert/README.md
new file mode 100644
index 0000000000..56898861f3
--- /dev/null
+++ b/model_cards/pvl/labse_bert/README.md
@@ -0,0 +1,47 @@
+---
+language: en
+thumbnail:
+tags:
+- bert
+- embeddings
+license: Apache-2.0
+---
+
+# LABSE BERT
+
+## Model description
+
+Model for "Language-agnostic BERT Sentence Embedding" paper from Fangxiaoyu Feng, Yinfei Yang, Daniel Cer, Naveen Arivazhagan, Wei Wang. Model available in [TensorFlow Hub](https://tfhub.dev/google/LaBSE/1).
+
+## Intended uses & limitations
+
+#### How to use
+
+```python
+from transformers import AutoTokenizer, AutoModel
+import torch
+
+# from sentence-transformers
+def mean_pooling(model_output, attention_mask):
+    token_embeddings = model_output[0] #First element of model_output contains all token embeddings
+    input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
+    sum_embeddings = torch.sum(token_embeddings * input_mask_expanded, 1)
+    sum_mask = torch.clamp(input_mask_expanded.sum(1), min=1e-9)
+    return sum_embeddings / sum_mask
+
+tokenizer = AutoTokenizer.from_pretrained("pvl/labse_bert", do_lower_case=False)
+model = AutoModel.from_pretrained("pvl/labse_bert")
+
+sentences = ['This framework generates embeddings for each input sentence',
+             'Sentences are passed as a list of string.',
+             'The quick brown fox jumps over the lazy dog.']
+
+encoded_input = tokenizer(sentences, padding=True, truncation=True, max_length=128, return_tensors='pt')
+
+with torch.no_grad():
+    model_output = model(**encoded_input)
+
+sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])
+
+
+```