antoinelouis commited on
Commit
b8fcf2a
1 Parent(s): d0fbe54

Upload folder using huggingface_hub

Browse files
1_Pooling/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "word_embedding_dimension": 768,
3
+ "pooling_mode_cls_token": false,
4
+ "pooling_mode_mean_tokens": true,
5
+ "pooling_mode_max_tokens": false,
6
+ "pooling_mode_mean_sqrt_len_tokens": false,
7
+ "pooling_mode_weightedmean_tokens": false,
8
+ "pooling_mode_lasttoken": false
9
+ }
README.md ADDED
@@ -0,0 +1,171 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ pipeline_tag: sentence-similarity
3
+ language: fr
4
+ license: apache-2.0
5
+ datasets:
6
+ - unicamp-dl/mmarco
7
+ metrics:
8
+ - recall
9
+ - posicube/mean_reciprocal_ranktags:
10
+
11
+ tags:
12
+ - sentence-transformers
13
+ - feature-extraction
14
+ - sentence-similarity
15
+ - transformers
16
+
17
+ ---
18
+
19
+ # biencoder-electra-base-french-europeana-cased-discriminator-mmarcoFR
20
+
21
+ 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. The model was trained on the **French** portion of the [mMARCO](https://huggingface.co/datasets/unicamp-dl/mmarco) dataset.
22
+
23
+ ## Usage
24
+ ***
25
+
26
+ #### Sentence-Transformers
27
+
28
+ Using this model becomes easy when you have [sentence-transformers](https://www.SBERT.net) installed:
29
+
30
+ ```
31
+ pip install -U sentence-transformers
32
+ ```
33
+
34
+ Then you can use the model like this:
35
+
36
+ ```python
37
+ from sentence_transformers import SentenceTransformer
38
+ sentences = ["This is an example sentence", "Each sentence is converted"]
39
+
40
+ model = SentenceTransformer('antoinelouis/biencoder-electra-base-french-europeana-cased-discriminator-mmarcoFR')
41
+ embeddings = model.encode(sentences)
42
+ print(embeddings)
43
+ ```
44
+
45
+
46
+
47
+ #### 🤗 Transformers
48
+
49
+ 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.
50
+
51
+ ```python
52
+ from transformers import AutoTokenizer, AutoModel
53
+ import torch
54
+
55
+
56
+ #Mean Pooling - Take attention mask into account for correct averaging
57
+ def mean_pooling(model_output, attention_mask):
58
+ token_embeddings = model_output[0] #First element of model_output contains all token embeddings
59
+ input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
60
+ return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)
61
+
62
+
63
+ # Sentences we want sentence embeddings for
64
+ sentences = ['This is an example sentence', 'Each sentence is converted']
65
+
66
+ # Load model from HuggingFace Hub
67
+ tokenizer = AutoTokenizer.from_pretrained('antoinelouis/biencoder-electra-base-french-europeana-cased-discriminator-mmarcoFR')
68
+ model = AutoModel.from_pretrained('antoinelouis/biencoder-electra-base-french-europeana-cased-discriminator-mmarcoFR')
69
+
70
+ # Tokenize sentences
71
+ encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')
72
+
73
+ # Compute token embeddings
74
+ with torch.no_grad():
75
+ model_output = model(**encoded_input)
76
+
77
+ # Perform pooling. In this case, mean pooling.
78
+ sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])
79
+
80
+ print("Sentence embeddings:")
81
+ print(sentence_embeddings)
82
+ ```
83
+
84
+
85
+
86
+ ## Evaluation
87
+ ***
88
+
89
+
90
+
91
+ We evaluated our model on the smaller development set of mMARCO-fr, which consists of 6,980 queries for a corpus of 8.8M candidate passages.
92
+ | MRR@10 | NDCG@10 | MAP@10 | Recall@10 | Recall@100 | Recall@500 |
93
+ |---------:|----------:|---------:|------------:|-------------:|-------------:|
94
+ | 23.38 | 27.97 | 22.91 | 43.5 | 68.96 | 81.61 |
95
+
96
+
97
+ Below, we compared its results with other biencoder models fine-tuned on the same dataset:
98
+ | | model | MRR@10 | NDCG@10 | MAP@10 | Recall@10 | Recall@100 (↑) | Recall@500 |
99
+ |---:|:-----------------------------------------------------------------------------------------------------------------------------------------------------------------------|---------:|----------:|---------:|------------:|-------------:|-------------:|
100
+ | 0 | [biencoder-camembert-base-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-camembert-base-mmarcoFR) | 28.53 | 33.72 | 27.93 | 51.46 | 77.82 | 89.13 |
101
+ | 1 | [biencoder-all-mpnet-base-v2-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-all-mpnet-base-v2-mmarcoFR) | 28.04 | 33.28 | 27.5 | 51.07 | 77.68 | 88.67 |
102
+ | 2 | [biencoder-multi-qa-mpnet-base-cos-v1-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-multi-qa-mpnet-base-cos-v1-mmarcoFR) | 27.6 | 32.92 | 27.09 | 50.97 | 77.41 | 87.79 |
103
+ | 3 | [biencoder-sentence-camembert-base-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-sentence-camembert-base-mmarcoFR) | 27.63 | 32.7 | 27.01 | 50.1 | 76.85 | 88.73 |
104
+ | 4 | [biencoder-distilcamembert-base-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-distilcamembert-base-mmarcoFR) | 26.8 | 31.87 | 26.23 | 49.2 | 76.44 | 87.87 |
105
+ | 5 | [biencoder-mpnet-base-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-mpnet-base-mmarcoFR) | 27.2 | 32.22 | 26.63 | 49.41 | 75.71 | 86.88 |
106
+ | 6 | [biencoder-multi-qa-distilbert-cos-v1-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-multi-qa-distilbert-cos-v1-mmarcoFR) | 26.36 | 31.26 | 25.82 | 47.93 | 75.42 | 86.78 |
107
+ | 7 | [biencoder-bert-base-uncased-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-bert-base-uncased-mmarcoFR) | 26.3 | 31.14 | 25.74 | 47.67 | 74.57 | 86.33 |
108
+ | 8 | [biencoder-msmarco-distilbert-cos-v5-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-msmarco-distilbert-cos-v5-mmarcoFR) | 25.75 | 30.63 | 25.24 | 47.22 | 73.96 | 85.64 |
109
+ | 9 | [biencoder-all-distilroberta-v1-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-all-distilroberta-v1-mmarcoFR) | 26.17 | 30.91 | 25.67 | 47.06 | 73.5 | 85.69 |
110
+ | 10 | [biencoder-all-MiniLM-L6-v2-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-all-MiniLM-L6-v2-mmarcoFR) | 25.49 | 30.39 | 24.99 | 47.1 | 73.48 | 86.09 |
111
+ | 11 | [biencoder-distilbert-base-uncased-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-distilbert-base-uncased-mmarcoFR) | 25.18 | 29.83 | 24.64 | 45.77 | 73.16 | 85.13 |
112
+ | 12 | [biencoder-msmarco-MiniLM-L12-cos-v5-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-msmarco-MiniLM-L12-cos-v5-mmarcoFR) | 26.22 | 30.99 | 25.69 | 47.29 | 73.09 | 84.95 |
113
+ | 13 | [biencoder-roberta-base-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-roberta-base-mmarcoFR) | 25.94 | 30.72 | 25.43 | 46.98 | 73.07 | 84.76 |
114
+ | 14 | [biencoder-distiluse-base-multilingual-cased-v1-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-distiluse-base-multilingual-cased-v1-mmarcoFR) | 24.57 | 29.08 | 24.04 | 44.51 | 72.54 | 85.13 |
115
+ | 15 | [biencoder-multi-qa-MiniLM-L6-cos-v1-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-multi-qa-MiniLM-L6-cos-v1-mmarcoFR) | 24.72 | 29.58 | 24.25 | 46.05 | 72.19 | 84.6 |
116
+ | 16 | [biencoder-MiniLM-L12-H384-uncased-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-MiniLM-L12-H384-uncased-mmarcoFR) | 25.43 | 30.1 | 24.88 | 46.13 | 72.16 | 83.84 |
117
+ | 17 | [biencoder-mMiniLMv2-L12-H384-distilled-from-XLMR-Large-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-mMiniLMv2-L12-H384-distilled-from-XLMR-Large-mmarcoFR) | 24.74 | 29.41 | 24.23 | 45.4 | 71.52 | 84.42 |
118
+ | 18 | [biencoder-electra-base-discriminator-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-electra-base-discriminator-mmarcoFR) | 24.77 | 29.37 | 24.21 | 45.2 | 70.84 | 83.25 |
119
+ | 19 | [biencoder-bert-medium-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-bert-medium-mmarcoFR) | 23.86 | 28.56 | 23.39 | 44.47 | 70.57 | 83.58 |
120
+ | 20 | [biencoder-msmarco-MiniLM-L6-cos-v5-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-msmarco-MiniLM-L6-cos-v5-mmarcoFR) | 24.39 | 28.96 | 23.91 | 44.58 | 70.36 | 82.88 |
121
+ | 21 | [biencoder-distilroberta-base-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-distilroberta-base-mmarcoFR) | 23.94 | 28.44 | 23.46 | 43.77 | 70.08 | 82.86 |
122
+ | 22 | [biencoder-camemberta-base-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-camemberta-base-mmarcoFR) | 24.78 | 29.24 | 24.23 | 44.58 | 69.59 | 82.18 |
123
+ | 23 | **biencoder-electra-base-french-europeana-cased-discriminator-mmarcoFR** | 23.38 | 27.97 | 22.91 | 43.5 | 68.96 | 81.61 |
124
+ | 24 | [biencoder-bert-small-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-bert-small-mmarcoFR) | 22.4 | 26.84 | 21.95 | 41.96 | 68.88 | 82.14 |
125
+ | 25 | [biencoder-mMiniLM-L6-v2-mmarcoFR-v2-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-mMiniLM-L6-v2-mmarcoFR-v2-mmarcoFR) | 22.87 | 27.26 | 22.37 | 42.3 | 68.78 | 81.39 |
126
+ | 26 | [biencoder-MiniLM-L6-H384-uncased-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-MiniLM-L6-H384-uncased-mmarcoFR) | 22.86 | 27.34 | 22.41 | 42.62 | 68.4 | 81.54 |
127
+ | 27 | [biencoder-deberta-v3-small-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-deberta-v3-small-mmarcoFR) | 22.44 | 26.84 | 21.97 | 41.84 | 68.17 | 80.9 |
128
+ | 28 | [biencoder-mMiniLMv2-L6-H384-distilled-from-XLMR-Large-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-mMiniLMv2-L6-H384-distilled-from-XLMR-Large-mmarcoFR) | 22.29 | 26.57 | 21.8 | 41.25 | 66.78 | 79.83 |
129
+ | 29 | [biencoder-bert-mini-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-bert-mini-mmarcoFR) | 20.06 | 24.09 | 19.66 | 37.78 | 64.27 | 77.39 |
130
+ | 30 | [biencoder-electra-small-discriminator-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-electra-small-discriminator-mmarcoFR) | 20.32 | 24.36 | 19.9 | 38.16 | 63.98 | 77.23 |
131
+ | 31 | [biencoder-deberta-v3-xsmall-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-deberta-v3-xsmall-mmarcoFR) | 17.7 | 21.29 | 17.31 | 33.59 | 58.76 | 73.45 |
132
+ | 32 | [biencoder-bert-tiny-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-bert-tiny-mmarcoFR) | 14.94 | 18.22 | 14.59 | 29.46 | 51.94 | 66.3 |
133
+ | 33 | [biencoder-t5-small-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-t5-small-mmarcoFR) | 12.44 | 15.1 | 12.14 | 24.28 | 47.82 | 63.37 |
134
+ | 34 | [biencoder-bert-small-mmarcoFR](https://huggingface.co/antoinelouis/biencoder-bert-small-mmarcoFR) | 0.22 | 0.28 | 0.21 | 0.5 | 1.25 | 2.34 |
135
+
136
+
137
+
138
+ ## Training
139
+ ***
140
+
141
+ #### Background
142
+
143
+ We used the [dbmdz/electra-base-french-europeana-cased-discriminator](https://huggingface.co/dbmdz/electra-base-french-europeana-cased-discriminator) model and fine-tuned it on a 500K sentence pairs dataset in French. We used 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. Formally, we compute the cos similarity from each possible sentence pairs from the batch. We then apply the cross entropy loss with a temperature of 0.05 by comparing with true pairs.
144
+
145
+ #### Hyperparameters
146
+
147
+ We trained the model on a single Tesla V100 GPU with 32GBs of memory during 20 epochs (i.e., 62.4k steps) using a batch size of 160. We used the AdamW optimizer with an initial learning rate of 2e-05, weight decay of 0.01, learning rate warmup over the first 500 steps, and linear decay of the learning rate. The sequence length was limited to 128 tokens.
148
+
149
+ #### Data
150
+
151
+ We used the French version of the [mMARCO](https://huggingface.co/datasets/unicamp-dl/mmarco) dataset to fine-tune our model. mMARCO is a multi-lingual machine-translated version of the MS MARCO dataset, a large-scale IR dataset comprising:
152
+ - a corpus of 8.8M passages;
153
+ - a training set of ~533k queries (with at least one relevant passage);
154
+ - a development set of ~101k queries;
155
+ - a smaller dev set of 6,980 queries (which is actually used for evaluation in most published works).
156
+ Link: [https://ir-datasets.com/mmarco.html#mmarco/v2/fr/](https://ir-datasets.com/mmarco.html#mmarco/v2/fr/)
157
+
158
+
159
+
160
+ ## Citation
161
+
162
+ ```bibtex
163
+ @online{louis2023,
164
+ author = 'Antoine Louis',
165
+ title = 'biencoder-electra-base-french-europeana-cased-discriminator-mmarcoFR: A Biencoder Model Trained on French mMARCO',
166
+ publisher = 'Hugging Face',
167
+ month = 'may',
168
+ year = '2023',
169
+ url = 'https://huggingface.co/antoinelouis/biencoder-electra-base-french-europeana-cased-discriminator-mmarcoFR',
170
+ }
171
+ ```
config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "dbmdz/electra-base-french-europeana-cased-discriminator",
3
+ "architectures": [
4
+ "ElectraModel"
5
+ ],
6
+ "attention_probs_dropout_prob": 0.1,
7
+ "classifier_dropout": null,
8
+ "embedding_size": 768,
9
+ "hidden_act": "gelu",
10
+ "hidden_dropout_prob": 0.1,
11
+ "hidden_size": 768,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 3072,
14
+ "layer_norm_eps": 1e-12,
15
+ "max_position_embeddings": 512,
16
+ "model_type": "electra",
17
+ "num_attention_heads": 12,
18
+ "num_hidden_layers": 12,
19
+ "pad_token_id": 0,
20
+ "position_embedding_type": "absolute",
21
+ "summary_activation": "gelu",
22
+ "summary_last_dropout": 0.1,
23
+ "summary_type": "first",
24
+ "summary_use_proj": true,
25
+ "torch_dtype": "float32",
26
+ "transformers_version": "4.26.1",
27
+ "type_vocab_size": 2,
28
+ "use_cache": true,
29
+ "vocab_size": 32000
30
+ }
config_sentence_transformers.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "__version__": {
3
+ "sentence_transformers": "2.2.2",
4
+ "transformers": "4.26.1",
5
+ "pytorch": "2.1.0.dev20230321+cu117"
6
+ }
7
+ }
dev_scores.csv ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ MRR@10,NDCG@10,MAP@10,Recall@10,Recall@100,Recall@500,model
2
+ 23.38,27.97,22.91,43.50,68.96,81.61,biencoder-electra-base-french-europeana-cased-discriminator-mmarcoFR
modules.json ADDED
@@ -0,0 +1,14 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [
2
+ {
3
+ "idx": 0,
4
+ "name": "0",
5
+ "path": "",
6
+ "type": "sentence_transformers.models.Transformer"
7
+ },
8
+ {
9
+ "idx": 1,
10
+ "name": "1",
11
+ "path": "1_Pooling",
12
+ "type": "sentence_transformers.models.Pooling"
13
+ }
14
+ ]
pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:19c1a49347204358f56d6592d53bc1e4521a2ddeae3b47620066969f400b249f
3
+ size 440177573
sentence_bert_config.json ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ {
2
+ "max_seq_length": 128,
3
+ "do_lower_case": false
4
+ }
special_tokens_map.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "cls_token": "[CLS]",
3
+ "mask_token": "[MASK]",
4
+ "pad_token": "[PAD]",
5
+ "sep_token": "[SEP]",
6
+ "unk_token": "[UNK]"
7
+ }
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1,17 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "cls_token": "[CLS]",
3
+ "do_basic_tokenize": true,
4
+ "do_lower_case": false,
5
+ "mask_token": "[MASK]",
6
+ "max_len": 512,
7
+ "model_max_length": 128,
8
+ "name_or_path": "dbmdz/electra-base-french-europeana-cased-discriminator",
9
+ "never_split": null,
10
+ "pad_token": "[PAD]",
11
+ "sep_token": "[SEP]",
12
+ "special_tokens_map_file": null,
13
+ "strip_accents": null,
14
+ "tokenize_chinese_chars": true,
15
+ "tokenizer_class": "ElectraTokenizer",
16
+ "unk_token": "[UNK]"
17
+ }
vocab.txt ADDED
The diff for this file is too large to render. See raw diff