Hugging Face's logo

almanach
/
camembertv2-base-rhapsodie

Token Classification
French
roberta
Eval Results
Model card Files Community
wissamantoun commited on
Commit
68d8e7d
•
1 Parent(s): cdba6eb

Upload folder using huggingface_hub

Browse files
Files changed (16) hide show
  1. README.md +161 -0
  2. camembertv2_base_p2_17k_last_layer.yaml +32 -0
  3. fr_rhapsodie-ud-dev.parsed.conllu +0 -0
  4. fr_rhapsodie-ud-test.parsed.conllu +0 -0
  5. model/config.json +1 -0
  6. model/lexers/camembertv2_base_p2_17k_last_layer/config.json +1 -0
  7. model/lexers/camembertv2_base_p2_17k_last_layer/model/config.json +30 -0
  8. model/lexers/camembertv2_base_p2_17k_last_layer/model/special_tokens_map.json +51 -0
  9. model/lexers/camembertv2_base_p2_17k_last_layer/model/tokenizer.json +0 -0
  10. model/lexers/camembertv2_base_p2_17k_last_layer/model/tokenizer_config.json +57 -0
  11. model/lexers/char_level_embeddings/config.json +1 -0
  12. model/lexers/fasttext/config.json +1 -0
  13. model/lexers/fasttext/fasttext_model.bin +3 -0
  14. model/lexers/word_embeddings/config.json +1 -0
  15. model/weights.pt +3 -0
  16. train.log +120 -0
README.md ADDED
@@ -0,0 +1,161 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language: fr
3
+ license: mit
4
+ tags:
5
+ - roberta
6
+ - token-classification
7
+ base_model: almanach/camembertv2-base
8
+ datasets:
9
+ - Rhapsodie
10
+ metrics:
11
+ - las
12
+ - upos
13
+ model-index:
14
+ - name: almanach/camembertv2-base-rhapsodie
15
+ results:
16
+ - task:
17
+ type: token-classification
18
+ name: Part-of-Speech Tagging
19
+ dataset:
20
+ type: Rhapsodie
21
+ name: Rhapsodie
22
+ metrics:
23
+ - name: upos
24
+ type: upos
25
+ value: 0.97556
26
+ verified: false
27
+ - task:
28
+ type: token-classification
29
+ name: Dependency Parsing
30
+ dataset:
31
+ type: Rhapsodie
32
+ name: Rhapsodie
33
+ metrics:
34
+ - name: las
35
+ type: las
36
+ value: 0.84497
37
+ verified: false
38
+ ---
39
+
40
+ # Model Card for almanach/camembertv2-base-rhapsodie
41
+
42
+ almanach/camembertv2-base-rhapsodie is a roberta model for token classification. It is trained on the Rhapsodie dataset for the task of Part-of-Speech Tagging and Dependency Parsing.
43
+ The model achieves an f1 score of on the Rhapsodie dataset.
44
+
45
+ The model is part of the almanach/camembertv2-base family of model finetunes.
46
+
47
+ ## Model Details
48
+
49
+ ### Model Description
50
+
51
+ - **Developed by:** Wissam Antoun (Phd Student at Almanach, Inria-Paris)
52
+ - **Model type:** roberta
53
+ - **Language(s) (NLP):** French
54
+ - **License:** MIT
55
+ - **Finetuned from model :** almanach/camembertv2-base
56
+
57
+ ### Model Sources
58
+
59
+ <!-- Provide the basic links for the model. -->
60
+
61
+ - **Repository:** https://github.com/WissamAntoun/camemberta
62
+ - **Paper:** https://arxiv.org/abs/2411.08868
63
+
64
+ ## Uses
65
+
66
+ The model can be used for token classification tasks in French for Part-of-Speech Tagging and Dependency Parsing.
67
+
68
+ ## Bias, Risks, and Limitations
69
+
70
+ The model may exhibit biases based on the training data. The model may not generalize well to other datasets or tasks. The model may also have limitations in terms of the data it was trained on.
71
+
72
+
73
+ ## How to Get Started with the Model
74
+
75
+ You can use the models directly with the hopsparser library in server mode https://github.com/hopsparser/hopsparser/blob/main/docs/server.md
76
+
77
+
78
+ ## Training Details
79
+
80
+ ### Training Procedure
81
+
82
+ Model trained with the [hopsparser](https://github.com/hopsparser/hopsparser) library on the Rhapsodie dataset.
83
+
84
+
85
+ #### Training Hyperparameters
86
+
87
+ ```yml
88
+ # Layer dimensions
89
+ mlp_input: 1024
90
+ mlp_tag_hidden: 16
91
+ mlp_arc_hidden: 512
92
+ mlp_lab_hidden: 128
93
+ # Lexers
94
+ lexers:
95
+ - name: word_embeddings
96
+ type: words
97
+ embedding_size: 256
98
+ word_dropout: 0.5
99
+ - name: char_level_embeddings
100
+ type: chars_rnn
101
+ embedding_size: 64
102
+ lstm_output_size: 128
103
+ - name: fasttext
104
+ type: fasttext
105
+ - name: camembertv2_base_p2_17k_last_layer
106
+ type: bert
107
+ model: /scratch/camembertv2/runs/models/camembertv2-base-bf16/post/ckpt-p2-17000/pt/
108
+ layers: [11]
109
+ subwords_reduction: "mean"
110
+ # Training hyperparameters
111
+ encoder_dropout: 0.5
112
+ mlp_dropout: 0.5
113
+ batch_size: 8
114
+ epochs: 64
115
+ lr:
116
+ base: 0.00003
117
+ schedule:
118
+ shape: linear
119
+ warmup_steps: 100
120
+
121
+ ```
122
+
123
+ #### Results
124
+
125
+ **UPOS:** 0.97556
126
+ **LAS:** 0.84497
127
+
128
+ ## Technical Specifications
129
+
130
+ ### Model Architecture and Objective
131
+
132
+ roberta custom model for token classification.
133
+
134
+ ## Citation
135
+
136
+ **BibTeX:**
137
+
138
+ ```bibtex
139
+ @misc{antoun2024camembert20smarterfrench,
140
+ title={CamemBERT 2.0: A Smarter French Language Model Aged to Perfection},
141
+ author={Wissam Antoun and Francis Kulumba and Rian Touchent and Éric de la Clergerie and Benoît Sagot and Djamé Seddah},
142
+ year={2024},
143
+ eprint={2411.08868},
144
+ archivePrefix={arXiv},
145
+ primaryClass={cs.CL},
146
+ url={https://arxiv.org/abs/2411.08868},
147
+ }
148
+
149
+ @inproceedings{grobol:hal-03223424,
150
+ title = {Analyse en dépendances du français avec des plongements contextualisés},
151
+ author = {Grobol, Loïc and Crabbé, Benoît},
152
+ url = {https://hal.archives-ouvertes.fr/hal-03223424},
153
+ booktitle = {Actes de la 28ème Conférence sur le Traitement Automatique des Langues Naturelles},
154
+ eventtitle = {TALN-RÉCITAL 2021},
155
+ venue = {Lille, France},
156
+ pdf = {https://hal.archives-ouvertes.fr/hal-03223424/file/HOPS_final.pdf},
157
+ hal_id = {hal-03223424},
158
+ hal_version = {v1},
159
+ }
160
+
161
+ ```
camembertv2_base_p2_17k_last_layer.yaml ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Layer dimensions
2
+ mlp_input: 1024
3
+ mlp_tag_hidden: 16
4
+ mlp_arc_hidden: 512
5
+ mlp_lab_hidden: 128
6
+ # Lexers
7
+ lexers:
8
+ - name: word_embeddings
9
+ type: words
10
+ embedding_size: 256
11
+ word_dropout: 0.5
12
+ - name: char_level_embeddings
13
+ type: chars_rnn
14
+ embedding_size: 64
15
+ lstm_output_size: 128
16
+ - name: fasttext
17
+ type: fasttext
18
+ - name: camembertv2_base_p2_17k_last_layer
19
+ type: bert
20
+ model: /scratch/camembertv2/runs/models/camembertv2-base-bf16/post/ckpt-p2-17000/pt/
21
+ layers: [11]
22
+ subwords_reduction: "mean"
23
+ # Training hyperparameters
24
+ encoder_dropout: 0.5
25
+ mlp_dropout: 0.5
26
+ batch_size: 8
27
+ epochs: 64
28
+ lr:
29
+ base: 0.00003
30
+ schedule:
31
+ shape: linear
32
+ warmup_steps: 100
fr_rhapsodie-ud-dev.parsed.conllu ADDED
The diff for this file is too large to render. See raw diff
 
fr_rhapsodie-ud-test.parsed.conllu ADDED
The diff for this file is too large to render. See raw diff
 
model/config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"mlp_input": 1024, "mlp_tag_hidden": 16, "mlp_arc_hidden": 512, "mlp_lab_hidden": 128, "biased_biaffine": true, "default_batch_size": 8, "encoder_dropout": 0.5, "extra_annotations": {}, "labels": ["acl", "acl:relcl", "advcl", "advcl:cleft", "advmod", "amod", "appos", "aux", "aux:caus", "aux:pass", "case", "cc", "ccomp", "compound", "conj", "cop", "csubj", "csubj:pass", "dep", "dep:comp", "det", "discourse", "dislocated", "expl:subj", "fixed", "flat", "iobj", "mark", "nmod", "nmod:appos", "nsubj", "nsubj:caus", "nsubj:pass", "nummod", "obj", "obj:lvc", "obl:agent", "obl:arg", "obl:mod", "parataxis", "parataxis:insert", "parataxis:parenth", "punct", "reparandum", "root", "vocative", "xcomp"], "mlp_dropout": 0.5, "tagset": ["ADJ", "ADP", "ADV", "AUX", "CCONJ", "DET", "INTJ", "NOUN", "NUM", "PRON", "PROPN", "PUNCT", "SCONJ", "VERB", "X"], "lexers": {"word_embeddings": "words", "char_level_embeddings": "chars_rnn", "fasttext": "fasttext", "camembertv2_base_p2_17k_last_layer": "bert"}, "multitask_loss": "sum"}
model/lexers/camembertv2_base_p2_17k_last_layer/config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"layers": [11], "subwords_reduction": "mean", "weight_layers": false}
model/lexers/camembertv2_base_p2_17k_last_layer/model/config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "/scratch/camembertv2/runs/models/camembertv2-base-bf16/post/ckpt-p2-17000/pt/",
3
+ "architectures": [
4
+ "RobertaForMaskedLM"
5
+ ],
6
+ "attention_probs_dropout_prob": 0.1,
7
+ "bos_token_id": 1,
8
+ "classifier_dropout": null,
9
+ "embedding_size": 768,
10
+ "eos_token_id": 2,
11
+ "hidden_act": "gelu",
12
+ "hidden_dropout_prob": 0.1,
13
+ "hidden_size": 768,
14
+ "initializer_range": 0.02,
15
+ "intermediate_size": 3072,
16
+ "layer_norm_eps": 1e-07,
17
+ "max_position_embeddings": 1025,
18
+ "model_name": "camembertv2-base-bf16",
19
+ "model_type": "roberta",
20
+ "num_attention_heads": 12,
21
+ "num_hidden_layers": 12,
22
+ "pad_token_id": 0,
23
+ "position_biased_input": true,
24
+ "position_embedding_type": "absolute",
25
+ "torch_dtype": "float32",
26
+ "transformers_version": "4.44.2",
27
+ "type_vocab_size": 1,
28
+ "use_cache": true,
29
+ "vocab_size": 32768
30
+ }
model/lexers/camembertv2_base_p2_17k_last_layer/model/special_tokens_map.json ADDED
@@ -0,0 +1,51 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token": {
3
+ "content": "[CLS]",
4
+ "lstrip": false,
5
+ "normalized": false,
6
+ "rstrip": false,
7
+ "single_word": false
8
+ },
9
+ "cls_token": {
10
+ "content": "[CLS]",
11
+ "lstrip": false,
12
+ "normalized": false,
13
+ "rstrip": false,
14
+ "single_word": false
15
+ },
16
+ "eos_token": {
17
+ "content": "[SEP]",
18
+ "lstrip": false,
19
+ "normalized": false,
20
+ "rstrip": false,
21
+ "single_word": false
22
+ },
23
+ "mask_token": {
24
+ "content": "[MASK]",
25
+ "lstrip": false,
26
+ "normalized": false,
27
+ "rstrip": false,
28
+ "single_word": false
29
+ },
30
+ "pad_token": {
31
+ "content": "[PAD]",
32
+ "lstrip": false,
33
+ "normalized": false,
34
+ "rstrip": false,
35
+ "single_word": false
36
+ },
37
+ "sep_token": {
38
+ "content": "[SEP]",
39
+ "lstrip": false,
40
+ "normalized": false,
41
+ "rstrip": false,
42
+ "single_word": false
43
+ },
44
+ "unk_token": {
45
+ "content": "[UNK]",
46
+ "lstrip": false,
47
+ "normalized": false,
48
+ "rstrip": false,
49
+ "single_word": false
50
+ }
51
+ }
model/lexers/camembertv2_base_p2_17k_last_layer/model/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
model/lexers/camembertv2_base_p2_17k_last_layer/model/tokenizer_config.json ADDED
@@ -0,0 +1,57 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": true,
3
+ "added_tokens_decoder": {
4
+ "0": {
5
+ "content": "[PAD]",
6
+ "lstrip": false,
7
+ "normalized": false,
8
+ "rstrip": false,
9
+ "single_word": false,
10
+ "special": true
11
+ },
12
+ "1": {
13
+ "content": "[CLS]",
14
+ "lstrip": false,
15
+ "normalized": false,
16
+ "rstrip": false,
17
+ "single_word": false,
18
+ "special": true
19
+ },
20
+ "2": {
21
+ "content": "[SEP]",
22
+ "lstrip": false,
23
+ "normalized": false,
24
+ "rstrip": false,
25
+ "single_word": false,
26
+ "special": true
27
+ },
28
+ "3": {
29
+ "content": "[UNK]",
30
+ "lstrip": false,
31
+ "normalized": false,
32
+ "rstrip": false,
33
+ "single_word": false,
34
+ "special": true
35
+ },
36
+ "4": {
37
+ "content": "[MASK]",
38
+ "lstrip": false,
39
+ "normalized": false,
40
+ "rstrip": false,
41
+ "single_word": false,
42
+ "special": true
43
+ }
44
+ },
45
+ "bos_token": "[CLS]",
46
+ "clean_up_tokenization_spaces": true,
47
+ "cls_token": "[CLS]",
48
+ "eos_token": "[SEP]",
49
+ "errors": "replace",
50
+ "mask_token": "[MASK]",
51
+ "model_max_length": 1000000000000000019884624838656,
52
+ "pad_token": "[PAD]",
53
+ "sep_token": "[SEP]",
54
+ "tokenizer_class": "RobertaTokenizer",
55
+ "trim_offsets": true,
56
+ "unk_token": "[UNK]"
57
+ }
model/lexers/char_level_embeddings/config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"char_embeddings_dim": 64, "output_dim": 128, "special_tokens": ["<root>"], "charset": ["<pad>", "<special>", "!", "'", ",", "-", ".", ":", "?", "A", "B", "C", "D", "E", "F", "G", "H", "I", "J", "K", "L", "M", "N", "O", "P", "Q", "R", "S", "T", "U", "V", "W", "X", "Y", "Z", "a", "b", "c", "d", "e", "f", "g", "h", "i", "j", "k", "l", "m", "n", "o", "p", "q", "r", "s", "t", "u", "v", "w", "x", "y", "z", "~", "\u00c0", "\u00c9", "\u00e0", "\u00e2", "\u00e4", "\u00e7", "\u00e8", "\u00e9", "\u00ea", "\u00ee", "\u00ef", "\u00f4", "\u00f9", "\u00fb", "\u00fc", "\u0153", "\u2026"]}
model/lexers/fasttext/config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"special_tokens": ["<root>"]}
model/lexers/fasttext/fasttext_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2d2e2a1042edcb221dc03be1c1b292bbedd1532687d5b01ce97b8053eda49529
3
+ size 800313178
model/lexers/word_embeddings/config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"embeddings_dim": 256, "unk_word": "<unk>", "vocabulary": ["!", ",", "-ce", "-chose", "-il", "-ils", "-l\u00e0", "-moi", "-m\u00eame", "-t-il", "-vous", ".", ":", "<root>", "<unk>", "?", "A", "Abeba", "Abidal", "Addis", "Aden", "Afrique", "Agutte", "Ag\u00fcero", "Albert", "Aligre", "Alsace", "Alsace-Lorraine", "Al\u00e9sia", "Anelka", "Angleterre", "Antech", "Antoine", "Arabie", "Arago", "Argentin", "Argentins", "Avennes", "A~", "B", "Bachelard", "Barthes", "Bastille", "Belgique", "Bel~", "Besse", "Bixente", "Branly", "Brochet", "CCJ", "CNRS", "CRDT", "Cannes", "Carrisou", "Cavalaire", "Censier", "Chambre", "Champagne", "Champollion", "Champon", "Chancellerie", "Chaplin", "Charlie", "Charlot", "Chavant", "Chroniques", "Ch~", "Claudel", "Clauwaert", "Commerce", "Dati", "Dazs", "Derni\u00e8res", "Diarra", "Diderot", "Dominique", "Dubedout", "D\u00e9p\u00eache", "Eclair", "Edouard", "Emmanu~", "Erikson", "Ethiopie", "Eug\u00e8ne", "Europe", "Facebook", "Famibook", "Faubourg", "Ferran", "Figaro", "Fillon", "Fnac", "France", "Francis", "Fran\u00e7ais", "Fran\u00e7aises", "Fran\u00e7ois", "Gabi", "Gaga", "Gago", "Gallas", "Gambetta", "Gargia", "Gibert", "Gide", "Giesbert", "Gourcuff", "Grande", "Grenette", "Grenoble", "Guti\u00e9rrez", "Gu~", "G\u00e9n\u00e9ral", "Habitat", "Haruki", "Haute", "Heinze", "Henri", "Henry", "Hilton", "Homme", "Honneur", "Hubert", "Hugo", "H\u00e4agen", "I", "Industrie", "Inter", "Internet", "Jaur\u00e8s", "Jean", "Jean-Christophe", "Joffrin", "Journal", "Jules", "K", "Kafka", "Kenya", "Klein", "L", "L'", "La", "Lafayette", "Laurent", "Le", "Les", "Lesdigui\u00e8res", "Libre", "Lib\u00e9ration", "Lionel", "Loisirs", "Lyon", "L\u00e9gion", "MDE", "MacDo", "Maison", "Malraux", "Mandanda", "Maradona", "Marne", "Marsabit", "Marseille", "Mascherano", "Massimo", "Mauriac", "Maxi", "Messi", "Mex\u00e8s", "Midi", "Mitterrand", "Monfreid", "Mougeotte", "Mref", "Mulhouse", "Murakami", "M~", "Nairobi", "Nation", "Nef", "Nicolas", "Nietzsche", "Nord", "Notre-Dame", "Nouvelles", "Occident", "Olkaloo", "Ora~", "PCB", "Palais", "Papa", "Pardigon", "Paris", "Parisien", "Pascale", "Paul", "Peugeot", "Picpus", "Platon", "Premier", "Prisse", "Progr\u00e8s", "Proust", "Provence", "Quatre", "Quentin", "Rachida", "Reims", "Reuilly-Diderot", "Rey", "Rodriguez", "Roger", "Roland", "Rolex", "Royal", "R\u00e9publique", "Saint", "Saint-Germain", "Saint-Jean-de-Maurienne", "Sainte-Claire", "Sarce", "Sarkozy", "Sartre", "Saules", "Seine", "Sembat", "Shoah", "Small", "Smic", "Solf\u00e9rino", "Somalie", "Sorbonne", "Sor~", "Stendhal", "Steve", "Strasbourg", "S\u00e9gol\u00e8ne", "Tarantino", "Thierry", "UMP", "Union", "VIP", "Vallier", "Val\u00e9ry", "Vaucanson", "Victor", "Vict~", "Vincennes", "Voix", "Wittgenstein", "World", "XXX", "YYY", "Yves", "Z", "Zanetti", "Zep", "a", "abominable", "abord", "aborder", "absolument", "absorb\u00e9s", "abstentionnisme", "abstenu", "accepte", "acceptons", "accessible", "accident", "accomplir", "accord", "accorde", "accordent", "accrue", "acc\u00e8s", "acc\u00e9l\u00e8re", "acc\u00e9l\u00e9rer", "acharn\u00e9e", "achats", "achetait", "ach\u00e8tent", "ach\u00e8tes", "ach\u00e8ve", "acquis", "acte", "action", "activit\u00e9", "actualit\u00e9", "actuellement", "admettre", "administration", "adolescent", "adore", "adresser", "adversaire", "affaire", "affam\u00e9e", "affectation", "affirmation", "affirme", "affirmer", "africains", "agenda", "agit", "agr\u00e9able", "ah", "ai", "aide", "aider", "aid\u00e9", "aid\u00e9e", "ailleurs", "aime", "aim\u00e9", "air", "aisthesis", "ait", "albums", "aliments", "allais", "allait", "allant", "allante", "allemand", "aller", "allez", "allions", "all\u00e9e", "alors", "amateurs", "ambiance", "ambigu", "amenait", "amie", "amis", "amnistie", "amour", "am\u00e8ne", "am\u00e9ricain", "an", "anciens", "anecdote", "anglais", "angle", "anglo", "anime", "annexe", "anniversaire", "ann\u00e9e", "ann\u00e9es", "ans", "anti-litt\u00e9rature", "antiquit\u00e9", "aper\u00e7u", "appara\u00eet", "appareillage", "appareils", "apparemment", "appartements", "appartenez", "appartenir", "appartient", "appelait", "appeler", "appelle", "appellerai", "apprendre", "appris", "approche", "appr\u00e9hension", "appr\u00eate", "appuyer", "apr\u00e8s", "arbitre", "argent", "argentin", "argentine", "argentins", "armer", "arrach\u00e9", "arrive", "arrivent", "arriver", "arrives", "arrivez", "arriv\u00e9", "arriv\u00e9e", "arri~", "arri\u00e8re", "arri\u00e8re-grand-m\u00e8re", "arri\u00e8re-plan", "arr\u00eat", "arr\u00eate", "arr\u00eater", "art", "artisans", "artiste", "artistes", "artistique", "aspect", "aspects", "aspirations", "assauts", "assez", "association", "associations", "associer", "assurer", "atmosph\u00e9riquement", "attachement", "attaquants", "attaque", "attarderai", "attendez", "attends", "attention", "attroupe", "au-dessus", "aucune", "auditeurs", "augment\u00e9", "aujourd'hui", "aupr\u00e8s", "aura", "auraient", "aurais", "aurait", "aussi", "aus~", "automatiquement", "automatisme", "automatismes", "autorit\u00e9", "autour", "autre", "autrefois", "autrement", "autres", "autrichien", "au~", "avaient", "avais", "avait", "avanc\u00e9", "avanc\u00e9es", "avant", "avec", "avenir", "aventure", "avenue", "avertissement", "aveugles", "avez", "avoir", "avons", "avoue", "ayant", "ayez", "a~", "baguette", "baguettes", "bah", "balader", "baladeur", "balaye", "ballon", "bande", "banlieue", "banque", "baraquements", "bas\u00e9e", "battus", "beau", "beaucoup", "bel", "belles", "ben", "besoin", "bien", "biens", "bio", "bless\u00e9e", "bleu", "bling", "bois", "bon", "bonne", "bonnes", "bons", "boude", "bougez", "boulanger", "boulevard", "bouleverse", "boulot", "bourgeoise", "bourgeoisie", "bourrasque", "bout", "boutique", "bou\u00e9e", "bo\u00eete", "branch\u00e9", "branch\u00e9e", "branch\u00e9s", "bras", "brasseries", "bref", "brioches", "brouhaha", "brousse", "br\u00fblante", "bulles", "bureau", "bus", "but", "buteur", "b~", "b\u00e2timent", "b\u00e9n\u00e9ficient", "b\u00e9n\u00e9voles", "b\u00e9r\u00e9zina", "c'", "c'est-\u00e0-dire", "cabine", "cadre", "cadres", "caetera", "caf\u00e9s", "camion", "camionnette", "camp", "campagne", "cantonales", "cap", "capable", "capacit\u00e9", "capitales", "capter", "car", "caricature", "carrefour", "carri\u00e8re", "carton", "cartons", "cas", "casser", "cause", "ca~", "ce", "ceci", "cela", "celles", "celui", "cent", "centre", "centre-ville", "cents", "cercle", "cercles", "certain", "certaine", "certainement", "certaines", "certains", "ces", "cet", "cette", "ceux", "chacun", "chacune", "chaleur", "chaleureusement", "chambre", "champ", "chance", "change", "changeant", "changeants", "changement", "changements", "changer", "chang\u00e9", "chaque", "charg\u00e9s", "charni\u00e8re", "chass\u00e9s", "chats", "chaussures", "chef", "chemins", "cher", "cherche", "chercher", "chers", "chez", "chiffres", "choisi", "chose", "choses", "ch~", "ch\u00f4meurs", "ci", "ciment", "cinglant", "cinq", "cinquantaine", "cinquante", "cin\u00e9ma", "circonscription", "circuler", "citer", "citoyens", "civils", "clair", "classe", "clavier", "client\u00e8le", "client\u00e8les", "clonage", "club", "clubbers", "cl\u00e9s", "coalis\u00e9es", "code", "cognitive", "coh\u00e9sion", "coin", "coll\u00e8ges", "com", "combat", "combats", "combien", "commandait", "commanditaire", "comme", "commence", "commencerai", "commences", "commenc\u00e9", "comment", "commen\u00e7ait", "commercial", "commer\u00e7ants", "commettre", "commissaire", "commun", "communautaire", "communautaires", "communaut\u00e9", "communications", "communiquer", "compatriotes", "complexe", "compliqu\u00e9", "compl\u00e8tement", "comporte", "composante", "comprend", "comprendre", "comprends", "compris", "compte", "comp\u00e9n\u00e8trent", "concept", "conception", "concernent", "concitoyens", "concurrence", "conditions", "confiance", "confiant", "conflit", "conf\u00e9rence", "conf\u00e9rences", "conjuguer", "connais", "connaissaient", "connaissance", "connaissent", "connaissez", "connaissiez", "connaissons", "conna\u00eet", "conna\u00eetre", "connue", "conqu\u00eate", "conseillerais", "conseils", "conserve", "consid\u00e9rable", "consid\u00e9rer", "consid\u00e9r\u00e9es", "consiste", "consommateur", "constate", "construire", "construit", "cons\u00e9quent", "contemporain", "contemporains", "continue", "continuer", "continues", "continuez", "continu\u00e9", "conti~", "contourne", "contours", "contraire", "contrat", "contre", "contre-attaques", "contre-litt\u00e9rature", "contrer", "contresens", "contribue", "contribu\u00e9", "contr\u00e9", "contr\u00f4le", "convaincu", "convictions", "convulsions", "con\u00e7ue", "coordination", "corne", "corner", "corps", "correctement", "corrompront", "corromp~", "couleurs", "couloir", "coup", "couples", "coupure", "courant", "courir", "cours", "course", "courses", "court", "courte", "cousins", "couverte", "craignent", "criminels", "crise", "critique", "croire", "crois", "croiser", "crudit\u00e9", "cruellement", "crues", "cr\u00e9ation", "cr\u00e9er", "cuisine", "culture", "culturelle", "curieux", "cyber", "c~", "c\u00e9r\u00e9monies", "c\u00f4t\u00e9", "c\u00f4t\u00e9s", "c\u0153ur", "d'", "d'ailleurs", "d'autant", "dame", "dans", "date", "davantage", "de", "dehors", "demain", "demandant", "demande", "demand\u00e9", "demie", "depuis", "dep~", "dernier", "derniers", "derri\u00e8re", "des", "descend", "descends", "descriptif", "descriptions", "design", "dessin\u00e9e", "dessous", "destruction", "deux", "deuxi\u00e8me", "devant", "devants", "devanture", "devenait", "devenu", "devient", "devra", "devraient", "devrais", "de~", "diagonale", "dictatures", "difficile", "difficiles", "diffuser", "diff\u00e9rence", "diff\u00e9rentes", "diff\u00e9rents", "dignit\u00e9", "dimension", "dimensions", "dirais", "dire", "directe", "direction", "dirigeant", "diriger", "dis", "disaient", "disciple", "disciplin\u00e9", "discours", "dise", "disent", "disiez", "disons", "dispara\u00eetre", "disparu", "dispose", "distinction", "distinctions", "district", "dis~", "dit", "dites", "diverses", "diviser", "dix", "dix-huit", "dix-huiti\u00e8me", "dix-neuf", "dix-neuvi\u00e8me", "dix-septi\u00e8me", "docteur", "docteurs", "doctors", "dois", "doit", "doivent", "domaine", "dominait", "donc", "donne", "donner", "donnerons", "donn\u00e9", "donn\u00e9es", "dont", "dormaient", "dortoir", "doute", "douter", "douze", "droit", "droite", "droites", "droits", "du", "dur", "d~", "d\u00e8s", "d\u00e9bats", "d\u00e9but", "d\u00e9butant", "d\u00e9buts", "d\u00e9b~", "d\u00e9charger", "d\u00e9cha\u00eenement", "d\u00e9chiffrer", "d\u00e9cid\u00e9", "d\u00e9cid\u00e9ment", "d\u00e9cisions", "d\u00e9concentr\u00e9", "d\u00e9coration", "d\u00e9coratives", "d\u00e9cor\u00e9es", "d\u00e9crire", "d\u00e9fendre", "d\u00e9fense", "d\u00e9fensive", "d\u00e9ferlante", "d\u00e9figur\u00e9s", "d\u00e9finir", "d\u00e9finit", "d\u00e9finition", "d\u00e9fricher", "d\u00e9gage", "d\u00e9gagement", "d\u00e9garni", "d\u00e9gradent", "d\u00e9j\u00e0", "d\u00e9late", "d\u00e9licat", "d\u00e9lit", "d\u00e9mocratie", "d\u00e9mocratique", "d\u00e9montrer", "d\u00e9m\u00e9nagements", "d\u00e9nomm\u00e9", "d\u00e9nonce", "d\u00e9part", "d\u00e9pass\u00e9", "d\u00e9pass\u00e9s", "d\u00e9pend", "d\u00e9penses", "d\u00e9placements", "d\u00e9route", "d\u00e9sagr\u00e9gation", "d\u00e9saveu", "d\u00e9sert", "d\u00e9signe", "d\u00e9sir", "d\u00e9sormais", "d\u00e9tente", "d\u00e9truits", "d\u00e9velopp\u00e9e", "d\u00fb", "effectivement", "effet", "efforcerait", "eh", "elle", "elle-m\u00eame", "elles", "embarque", "emb\u00eatant", "emmen\u00e9", "emm\u00e9nagements", "empire", "emploi", "employ\u00e9", "emprise", "emprunte", "en", "encha\u00eene", "encore", "endeuill\u00e9es", "endormies", "endroit", "enfance", "enfant", "enfants", "enfin", "enfuit", "engagent", "enjeu", "ennui", "enseignants", "ensemble", "ensuite", "entendait", "entendre", "enthousiasme", "entraide", "entre", "entreprises", "entrer", "envie", "environ", "environnement", "envoy\u00e9", "erreur", "erreurs", "es", "esprit", "esp\u00e8ce", "esp\u00e9rance", "esp\u00e9rez", "essai", "essais", "essayer", "essentielles", "est", "esth\u00e9tique", "et", "eu", "euh", "eus", "eux", "exactement", "excellent", "exceptionnel", "except\u00e9", "exclamation", "excusez", "exemple", "exercice", "exerc\u00e9", "exigence", "exigences", "existe", "exist\u00e9", "explication", "explications", "explique", "expliquer", "exploser", "exprimer", "exp~", "exp\u00e9die", "exp\u00e9rience", "exp\u00e9rimente", "extraordinaire", "extr\u00eamement", "ex~", "ex\u00e9cutif", "fabrication", "fabrique", "fabriquer", "face", "facile", "faiblesse", "faire", "fais", "faisait", "faisant", "faisiez", "fait", "faites", "fallait", "fallu", "fameux", "familial", "famille", "familles", "fantaisie", "fantastique", "fassiez", "fatalit\u00e9", "faudrait", "fausse", "faut", "faute", "fautes", "fa\u00e7on", "femme", "femmes", "fen\u00eatres", "ferm\u00e9", "ferm\u00e9e", "ferm\u00e9s", "feront", "festival", "feux", "fff", "fid\u00e8le", "fiert\u00e9", "fig\u00e9", "fille", "fils", "fin", "finalement", "financier", "fini", "finissent", "fi\u00e8vre", "flagrant", "flics", "flou", "floue", "flying", "foi", "foire", "fois", "fonci\u00e8rement", "fonctions", "fond", "fondamentale", "font", "fontaine", "force", "forc\u00e9ment", "forme", "former", "formes", "formidable", "fort", "forte", "forts", "for\u00eats", "foss\u00e9", "foule", "fragilit\u00e9", "frais", "franc", "franchira", "franchit", "fran\u00e7ais", "fran\u00e7aise", "fran\u00e7aises", "frappe", "fraternel", "fraternelle", "friandises", "front", "fronti\u00e8re", "fr\u00e9quenter", "fut", "futur", "f~", "f\u00e9minins", "f\u00e9tiches", "gaillard", "galeries", "garantissent", "garantit", "garde", "gardons", "gare", "gar\u00e7on", "gauche", "gazon", "gens", "gifle", "globalement", "goulag", "gouvernement", "go\u00fbt\u00e9", "grand", "grande", "grandes", "grandir", "grands", "grands-parents", "gratuits", "grec", "gros", "gr\u00e8ve", "gr\u00eale", "guerre", "guerres", "guillemets", "g\u00e2teaux", "g\u00e9nie", "g\u00e9n\u00e9ral", "g\u00e9n\u00e9rale", "g\u00e9n\u00e9raliste", "g\u00e9n\u00e9ration", "g\u00e9n\u00e9rations", "g\u00e9n\u00e9raux", "g\u00e9n\u00e9tiques", "g\u00e9ographe", "habiller", "habill\u00e9e", "habit", "habitudes", "halle", "hasard", "haut", "haute", "hein", "heure", "heures", "heureuse", "heureusement", "heureux", "heurte", "heurtent", "hier", "histoire", "histoires", "historique", "historiques", "homme", "hommes", "honneur", "honteux", "horaire", "horaires", "horreur", "huit", "humaine", "humanit\u00e9", "humilit\u00e9", "humour", "h\u00e9donistes", "h\u00e9donistique", "h\u00e9do~", "h\u00e9ritiers", "h\u00e9sitations", "h\u00f4pital", "ici", "identit\u00e9", "id\u00e9e", "il", "ils", "image", "imagine", "immeuble", "immobilier", "imm\u00e9diat", "imm\u00e9diatement", "importance", "important", "importante", "importantes", "importe", "impose", "impression", "imprimantes", "inconscience", "incontournable", "incroyable", "indiqu\u00e9", "ind\u00e9boulonnable", "ind\u00e9pendant", "informaticien", "informatique", "informe", "infos", "initiative", "inquiets", "installent", "installer", "instant", "instants", "instaur\u00e9", "institution", "institutionnel", "institutions", "instruction", "intellectuels", "intenable", "international", "internet", "interpr\u00e9tations", "interroger", "intersection", "intervenir", "intervenu", "int\u00e9ressait", "int\u00e9ressant", "int\u00e9ressants", "int\u00e9ress\u00e9e", "int\u00e9rieur", "int\u00e9rieurs", "int\u00e9r\u00eat", "inventer", "invitation", "invitent", "invit\u00e9e", "irr\u00e9elles", "itin\u00e9raires", "j'", "jalonnent", "jamais", "jardin", "jardins", "je", "jet", "jeu", "jeune", "jeunes", "jeunesse", "jeu~", "jouaient", "joue", "jouer", "joueur", "joueurs", "jouissance", "jour", "journaux", "journ\u00e9e", "jours", "jou\u00e9", "joyeux", "jugement", "jusqu'", "juste", "justement", "justice", "j~", "kilom\u00e8tres", "kiosque", "l'", "l'on", "la", "laboratoire", "laisse", "langage", "langue", "laquelle", "larcin", "large", "largement", "le", "leaders", "leadership", "lecteur", "lecture", "lequel", "les", "leur", "leurs", "le\u00e7on", "libert\u00e9", "libert\u00e9s", "libre", "licenciement", "lieu", "ligne", "lignes", "limites", "lire", "lisais", "lisez", "lit", "litigieuse", "littoral", "litt\u00e9raire", "litt\u00e9ralement", "litt\u00e9rature", "litt\u00e9r~", "livraison", "livre", "livres", "li\u00e9", "li\u00e9e", "li\u00e9es", "li\u00e9s", "local", "location", "logiciels", "loin", "lointain", "loisir", "long", "longe", "longer", "longez", "longtemps", "lorgne", "lorsqu'", "lorsque", "louper", "loyers", "lu", "lui", "lui-m\u00eame", "lus", "lyc\u00e9e", "lyc\u00e9en", "lyc\u00e9es", "l~", "l\u00e0", "l\u00e0-bas", "l\u00e0-dedans", "l\u00e0-dessus", "l\u00e2chent", "l\u00e2cher", "l\u00e2ch\u00e9", "l\u00e8guerons", "l\u00e9g\u00e8ret\u00e9", "m'", "ma", "machin", "machine", "magasin", "magnifique", "main", "maintenance", "maintenant", "maintenan~", "maintenir", "maint~", "maires", "mais", "majorit\u00e9", "mal", "malade", "maladie", "maladive", "malheur", "malheureusement", "mang\u00e9", "manifestations", "manipulations", "mani\u00e8re", "manque", "manqu\u00e9", "man\u00e8ge", "man\u0153uvre", "marchandises", "marche", "march\u00e9", "mari\u00e9e", "marquage", "marque", "marqu\u00e9", "marrant", "masculins", "masques", "match", "matchs", "maternelles", "matin", "mati\u00e8re", "mat\u00e9rielles", "mauvais", "mauvaise", "ma~", "ma\u00eetrisait", "ma\u00eetris\u00e9e", "me", "meilleures", "membres", "men\u00e9e", "merci", "mes", "message", "mesure", "mesures", "mesurons", "mettre", "mh", "mi-temps", "micro-processeurs", "midi", "mien", "mieux", "milieu", "militaires", "mille", "millions", "mine", "ministre", "minute", "minutes", "miraculeux", "mise", "mises", "mobile", "mobilisation", "mobilis\u00e9s", "moderne", "modernit\u00e9", "modifier", "moi", "moins", "moment", "moments", "mon", "monde", "mondiales", "mondialisation", "monsieur", "monter", "montes", "montre", "montrer", "monuments", "morbide", "morceau", "mort", "mortels", "mor~", "mot", "motif", "motivations", "mourir", "mouton", "mouvements", "moyenne", "moyennes", "moyens", "mule", "mulot", "municipales", "musicale", "musique", "mus\u00e9e", "mythe", "m~", "m\u00e8ne", "m\u00e8nent", "m\u00e8tres", "m\u00e8~", "m\u00e9decin", "m\u00e9decine", "m\u00e9decins", "m\u00e9diane", "m\u00e9diateur", "m\u00e9lange", "m\u00e9moire", "m\u00e9tier", "m\u00e9tiers", "m\u00e9tro", "m\u00e9tropole", "m\u00e9t~", "m\u00e9~", "m\u00eale", "m\u00eame", "n'", "nation", "national", "nations", "naturel", "naturellement", "naturels", "naufrage", "na\u00eetre", "na\u00effs", "ne", "neuf", "ni", "niche", "nier", "niveau", "ni~", "noir", "nom", "nombre", "nombreuses", "nombreux", "nomination", "noms", "non", "nord", "normalement", "nos", "notamment", "notion", "notre", "nous", "nouveau", "nouveaux", "nouvel", "nouvelle", "nouvelles", "nuages", "nuanc\u00e9e", "nue", "nul", "num\u00e9rique", "n\u00e9anmoins", "n\u00e9cessaire", "n\u00e9e", "n\u00e9gation", "n\u00e9gociations", "n\u00e9s", "obligatoire", "oblig\u00e9e", "oblig\u00e9s", "obstacles", "obstinent", "obstruction", "obs\u00e8de", "occasion", "occupe", "occup\u00e9es", "offensif", "officiers", "oh", "oiseau", "on", "ont", "oppose", "opposer", "optimiste", "or", "ordinateur", "ordinateurs", "ordre", "ordres", "organiser", "organis\u00e9e", "orientent", "orient\u00e9es", "orient\u00e9s", "oser", "ou", "ouais", "oublie", "oubli\u00e9", "ouest", "ouh", "oui", "outrage", "outre-mer", "ouverte", "ouverture", "ouvrir", "o\u00f9", "page", "pain", "paix", "par", "paraissait", "para\u00eet", "para\u00eetre", "parc", "parce", "pardon", "parents", "parfois", "parisienne", "parlais", "parlant", "parle", "parler", "parlerai", "parlez", "parl\u00e9", "parmi", "part", "partageons", "partager", "partant", "partent", "partez", "parti", "participation", "participer", "particulier", "particuli\u00e8rement", "partie", "partiels", "partir", "partis", "pas", "passage", "passait", "passant", "passa~", "passe", "passent", "passer", "passerelle", "passes", "passez", "passion", "passionnant", "pass\u00e9", "patrimoine", "patron", "pauvre", "pauvret\u00e9", "pays", "peine", "peinture", "penalty", "pendant", "pensais", "pensait", "pensant", "pensante", "pense", "penser", "penseur", "pensez", "pensum", "pens\u00e9e", "pens\u00e9es", "pen~", "people", "peoples", "perdre", "perdu", "performance", "permanence", "permet", "permettent", "personnage", "personnages", "perte", "peser", "petit", "petite", "petites", "petits", "peu", "peuples", "peur", "peut", "peut-\u00eatre", "peut~", "peuvent", "peux", "pff", "philo", "philosophe", "philosophes", "philosophie", "philosophique", "photos", "phrase", "phrases", "ph\u00e9nom\u00e8ne", "pied", "pieds", "piquer", "piscine", "pi\u00e8ge", "place", "places", "plaisaient", "plaisir", "plant\u00e9", "plan\u00e8te", "plateau", "plateaux", "plein", "plong\u00e9", "plupart", "plus", "plusieurs", "plut\u00f4t", "point", "points", "poitrine", "polarisaient", "police", "policier", "policiers", "politique", "politiques", "pompiers", "ponctuel", "ponctuels", "porche", "portes", "pose", "poser", "posez", "possession", "poss\u00e9dait", "poste", "postes", "poteau", "pour", "pourquoi", "pourra", "pourrais", "pourrait", "pourriez", "poursuivez", "poursuivre", "pourtant", "pouss\u00e9", "pouvais", "pouvait", "pouvez", "pouvoir", "po\u00e9sie", "po\u00ef\u00e9sis", "pratiquant", "pratique", "pratiquement", "pratiquez", "pra~", "premier", "premi\u00e8re", "prend", "prendre", "prendront", "prends", "presqu'", "presque", "presse", "preuve", "primaire", "principal", "principalement", "priori", "pris", "prise", "privil\u00e9giait", "privil\u00e9gi\u00e9s", "priv\u00e9e", "prix", "probl\u00e8me", "probl\u00e8mes", "prochainement", "proches", "production", "productivit\u00e9", "produit", "professeurs", "profite", "profiter", "profond\u00e9ment", "programmation", "programme", "programmes", "progr\u00e8s", "projets", "prolonge", "prolongement", "promener", "pronostique", "propos", "proposons", "propos\u00e9s", "propre", "protestation", "province", "pr\u00e8s", "pr\u00e9alable", "pr\u00e9carit\u00e9", "pr\u00e9cieux", "pr\u00e9cipite", "pr\u00e9cise", "pr\u00e9cision", "pr\u00e9cis\u00e9ment", "pr\u00e9c\u00e9dent", "pr\u00e9d\u00e9cesseurs", "pr\u00e9fecture", "pr\u00e9fet", "pr\u00e9gnante", "pr\u00e9liminaires", "pr\u00e9occupent", "pr\u00e9pa", "pr\u00e9sent", "pr\u00e9sentes", "pr\u00e9sident", "pr\u00e9sidentielle", "pr\u00e9vient", "pr\u00e9voir", "pr\u00e9vu", "pu", "public", "publics", "publiques", "pudeur", "pudique", "puis", "puisqu'", "puisque", "puissamment", "puissance", "puisse", "p~", "p\u00e2tisserie", "p\u00e2tissier", "p\u00e8re", "p\u00e8sent", "p\u00e9riph\u00e9riques", "qu'", "quai", "quais", "quand", "quarante", "quarante-huit", "quarante-neuf", "quaranti\u00e8me", "quartier", "quartiers", "quar~", "quatorze", "quatorzi\u00e8me", "quatre", "quatre-vingt-dix-neuf", "quatre-vingt-onze", "quatre-vingt-six", "que", "quel", "quelle", "quelqu'un", "quelque", "quelquefois", "quelques", "question", "qui", "quinze", "quoi", "quoique", "quotidienne", "racont\u00e9", "rails", "raison", "rappelez", "rapport", "rapproche", "rassembl\u00e9e", "ratatin\u00e9", "rattraper", "rayon", "rebondit", "recevoir", "recherche", "recherch\u00e9e", "reconnaissez", "reconna\u00eet", "reconna\u00eetre", "records", "recueil", "reculer", "redis", "reflet", "refus\u00e9es", "regarde", "regarder", "regardez", "rejoindre", "rejoint", "relativement", "religieuse", "religieuses", "relire", "relis", "rel\u00e2chement", "rel\u00e8vement", "remarque", "remercie", "remis", "remontant", "remonte", "remonter", "remontez", "rempla\u00e7ante", "renaissance", "rencontre", "rencontrent", "rencontres", "rencontr\u00e9", "rencontr\u00e9s", "rendez", "rendre", "renoncement", "rentre", "rentrent", "rentrer", "rentres", "rentr\u00e9", "rentr\u00e9e", "rentr\u00e9s", "renverse", "repartie", "repartis", "repensant", "repos", "repousse", "repousser", "reprends", "reprenez", "repris", "repr\u00e9senter", "rep\u00e9rer", "respect", "respecter", "respectueuse", "responsabilit\u00e9", "responsabilit\u00e9s", "responsable", "responsables", "ressemble", "ressentie", "ressentir", "ressentons", "ressort", "restaient", "restant", "reste", "restent", "rest\u00e9e", "retombez", "retour", "retournent", "retourner", "retrait", "retrouve", "retrouver", "revenant", "revenir", "revenu", "revenue", "reviennent", "reviens", "revient", "re\u00e7oit", "rh\u00e9torique", "ricanement", "riche", "ridicule", "rien", "rivage", "romans", "romantique", "rond-point", "rose", "route", "rue", "rues", "rythme", "r\u00e8gle", "r\u00e8gles", "r\u00e9action", "r\u00e9aliser", "r\u00e9cemment", "r\u00e9cents", "r\u00e9ception", "r\u00e9ciproquement", "r\u00e9cup\u00e9rer", "r\u00e9cup\u00e9r\u00e9", "r\u00e9flexion", "r\u00e9forme", "r\u00e9formes", "r\u00e9f\u00e9rence", "r\u00e9gion", "r\u00e9gions", "r\u00e9gl\u00e9", "r\u00e9habiliter", "r\u00e9paration", "r\u00e9pond", "r\u00e9pondre", "r\u00e9ponse", "r\u00e9publicain", "r\u00e9seaux", "r\u00e9serv\u00e9", "r\u00e9sultat", "r\u00e9sumer", "r\u00e9tablir", "r\u00e9ussirons", "r\u00eaver", "r\u00eaveur", "r\u00f4le", "s'", "sa", "sacr\u00e9es", "sais", "sait", "salaires", "sanction", "sanctionn\u00e9e", "sandwich", "sans", "sant\u00e9", "sauf", "saura", "saute", "sauvage", "sauvagement", "sauve", "sauvetage", "savaient", "savez", "savoir", "savons", "sav~", "sa~", "science", "scientifique", "scolaire", "scolarit\u00e9", "scrutin", "sc\u00e8ne", "se", "second", "seconde", "seigneur", "seizi\u00e8me", "selon", "semaines", "semblait", "semblant", "semble", "semblerait", "semblez", "sens", "sensation", "sentiment", "sen~", "sept", "septi\u00e8me", "sera", "serait", "seras", "serez", "serons", "serveur", "service", "services", "ses", "set", "seul", "seule", "seulement", "sexe", "sexualit\u00e9", "se~", "si", "siffl\u00e9", "signale", "signes", "sillonnant", "simple", "simplement", "singulier", "sinon", "site", "sites", "situation", "situe", "situer", "sixi\u00e8me", "si\u00e8cle", "si\u00e8cles", "social", "sociale", "socialiste", "sociaux", "sociologie", "soci\u00e9t\u00e9", "soci\u00e9t\u00e9s", "soi", "soient", "soir", "soit", "soixante-cinq", "soixante-quatre", "soixante-quinze", "sol", "soleil", "solidaire", "solidarit\u00e9", "solidarit\u00e9s", "solide", "solution", "sommes", "son", "sont", "sors", "sort", "sorte", "sorti", "sortie", "sortir", "soucis", "souffrance", "souffrent", "souhaite", "soulignent", "souligner", "souris", "sous", "soutien", "souvenez", "souvenir", "souvenirs", "souvent", "souviens", "sovi\u00e9tique", "spectateur", "spontan\u00e9", "sport", "sp\u00e9cialement", "sp\u00e9cialiste", "sp\u00e9cialis\u00e9", "sp\u00e9cialis\u00e9s", "square", "stade", "stationner", "stations", "statut", "stopp\u00e9", "stressant", "su", "successives", "succ\u00e8s", "sud", "suffit", "suffrage", "suis", "suite", "suivez", "suivi", "suivre", "sujet", "sur", "surface", "surnaturelles", "surtout", "su~", "symbolis\u00e9", "sympathie", "syst\u00e8me", "s~", "s\u00e9curit\u00e9", "s\u00e9duction", "s\u00e9parable", "s\u00e9quence", "s\u00e9rieux", "s\u00fbr", "s\u00fbre", "s\u00fbrement", "tandis", "tant", "tard", "tchater", "techniquement", "tel", "telle", "tellement", "telles", "tels", "temps", "temp\u00eate", "tendance", "tendu", "tendue", "tenir", "tentent", "tenu", "terme", "termine", "terrasses", "terre", "territoriale", "test", "th\u00e9matiques", "th\u00e9oriciens", "th\u00e9orie", "th\u00e9oris\u00e9", "tienne", "tient", "tir", "tire", "tirer", "tissu", "toi", "tol\u00e9rance", "tombe", "tombent", "tomber", "tomb\u00e9e", "tontons", "tornade", "totalement", "totalitaires", "touche", "toucher", "touchons", "touch\u00e9s", "toujours", "tour", "tourisme", "tourments", "tournait", "tourne", "tourner", "tournes", "tournez", "tous", "tout", "toute", "toutes", "tou~", "trace", "traces", "tradition", "trag\u00e9dies", "train", "traiter", "traiteur", "trajet", "tram", "trams", "tranquille", "transform\u00e9", "transporte", "transports", "travail", "travaillais", "travaille", "travailler", "travaillez", "travaill\u00e9", "travaux", "travers", "traverse", "traverser", "traverses", "traversez", "treize", "trente", "trente-deux", "trente-et-un", "trente-neuf", "triomphe", "triste", "trois", "troisi\u00e8me", "tromper", "trop", "trottoir", "trou", "trouve", "trouver", "trouverai", "trouv\u00e9", "truc", "trucs", "tr\u00e8s", "tu", "type", "t~", "t\u00e2che", "t\u00e9l\u00e9phone", "t\u00e9l\u00e9ph\u00e9rique", "t\u00e9l\u00e9vis\u00e9", "t\u00eate", "t\u00f4t", "un", "une", "unie", "uniquement", "univers", "universel", "uns", "usine", "utile", "utilisateur", "utilisateurs", "utilise", "utiliser", "utilises", "utilis\u00e9", "u~", "va", "vacances", "vachement", "vague", "vais", "valeurs", "variations", "varie", "vari\u00e9", "vas", "vaut", "venait", "venez", "venir", "vente", "venue", "verra", "verras", "verrez", "verront", "vers", "version", "vert", "vertueuse", "veulent", "veut", "veux", "victimes", "victoire", "vid\u00e9os", "vie", "vieil", "vieille", "vieillissent", "viennent", "viens", "vient", "vif", "ville", "villes", "vingt", "vingt-cinq", "vingt-deux", "vingt-deuxi\u00e8me", "vingt-et-uni\u00e8me", "vingt-neuf", "vingt-trois", "vingt~", "visiblement", "visualise", "vite", "vitrine", "vitrines", "vivant", "vivante", "vivent", "vivons", "vivre", "voies", "voil\u00e0", "voir", "vois", "voisin", "voisins", "voit", "voiture", "voitures", "vol", "vole", "voler", "volontaire", "volont\u00e9", "vol\u00e9", "vont", "vos", "vote", "votre", "voudrais", "voudrait", "voudrons", "voulais", "voulait", "voulez", "vouloir", "voulu", "vous", "vous-m\u00eame", "voyaient", "voyait", "voyez", "voyons", "vrai", "vraiment", "vu", "vue", "vues", "v~", "v\u00e9cu", "v\u00e9ritable", "v\u00e9rit\u00e9", "y", "yeux", "ziki", "z\u00e9ro", "\u00c0", "\u00c9tat", "\u00c9tienne", "\u00c9toile", "\u00e0", "\u00e2g\u00e9e", "\u00e2me", "\u00e7a", "\u00e7~", "\u00e9bahie", "\u00e9bauche", "\u00e9carte", "\u00e9changer", "\u00e9chapper", "\u00e9cho", "\u00e9choient", "\u00e9clairages", "\u00e9clectique", "\u00e9cole", "\u00e9coles", "\u00e9crasante", "\u00e9crire", "\u00e9cris", "\u00e9crit", "\u00e9crite", "\u00e9criture", "\u00e9crivain", "\u00e9crivains", "\u00e9crivant", "\u00e9crivent", "\u00e9ditorial", "\u00e9ducation", "\u00e9galement", "\u00e9galisation", "\u00e9gards", "\u00e9glise", "\u00e9lan", "\u00e9lecteurs", "\u00e9lectorales", "\u00e9lectronique", "\u00e9loge", "\u00e9lus", "\u00e9l\u00e9ments", "\u00e9minemment", "\u00e9motion", "\u00e9mu", "\u00e9nergiquement", "\u00e9nerv\u00e9", "\u00e9ner~", "\u00e9norme", "\u00e9norm\u00e9ment", "\u00e9poque", "\u00e9poques", "\u00e9pouvantable", "\u00e9pouvantablement", "\u00e9preuve", "\u00e9prouve", "\u00e9prouver", "\u00e9puise", "\u00e9quateur", "\u00e9quipe", "\u00e9taient", "\u00e9tais", "\u00e9tait", "\u00e9tant", "\u00e9thiopien", "\u00e9thique", "\u00e9tiez", "\u00e9tonnement", "\u00e9trange", "\u00e9tranger", "\u00e9trangers", "\u00e9tudes", "\u00e9tudiant", "\u00e9tu~", "\u00e9t~", "\u00e9t\u00e9", "\u00e9videmment", "\u00e9vidence", "\u00e9vident", "\u00e9viter", "\u00e9volution", "\u00e9voqu\u00e9", "\u00e9v\u00e8nements", "\u00e9~", "\u00eates", "\u00eatre", "\u00eele", "\u0153uvre", "\u0153uvres", "\u2026"], "word_dropout": 0.5}
model/weights.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f52e4cda95f16f9a7c591c0d99e480045a895280dbb6d9c16c29ad3c7823fc57
3
+ size 1743252842
train.log ADDED
@@ -0,0 +1,120 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [hops] 2024-09-23 16:13:21.055 | INFO | Initializing a parser from /workspace/configs/exp_camembertv2/camembertv2_base_p2_17k_last_layer.yaml
2
+ [hops] 2024-09-23 16:13:21.210 | INFO | Generating a FastText model from the treebank
3
+ [hops] 2024-09-23 16:13:21.214 | INFO | Training fasttext model
4
+ [hops] 2024-09-23 16:13:22.656 | WARNING | Some weights of RobertaModel were not initialized from the model checkpoint at /scratch/camembertv2/runs/models/camembertv2-base-bf16/post/ckpt-p2-17000/pt/ and are newly initialized: ['roberta.pooler.dense.bias', 'roberta.pooler.dense.weight']
5
+ You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
6
+ [hops] 2024-09-23 16:13:34.964 | INFO | Start training on cuda:3
7
+ [hops] 2024-09-23 16:13:35.203 | WARNING | You're using a RobertaTokenizerFast tokenizer. Please note that with a fast tokenizer, using the `__call__` method is faster than using a method to encode the text followed by a call to the `pad` method to get a padded encoding.
8
+ [hops] 2024-09-23 16:13:59.790 | INFO | Epoch 0: train loss 3.0280 dev loss 2.5852 dev tag acc 18.31% dev head acc 16.94% dev deprel acc 32.20%
9
+ [hops] 2024-09-23 16:13:59.791 | INFO | New best model: head accuracy 16.94% > 0.00%
10
+ [hops] 2024-09-23 16:14:09.299 | INFO | Epoch 1: train loss 2.3355 dev loss 1.8455 dev tag acc 43.37% dev head acc 38.18% dev deprel acc 54.33%
11
+ [hops] 2024-09-23 16:14:09.300 | INFO | New best model: head accuracy 38.18% > 16.94%
12
+ [hops] 2024-09-23 16:14:19.117 | INFO | Epoch 2: train loss 1.8050 dev loss 1.4588 dev tag acc 48.38% dev head acc 54.01% dev deprel acc 63.87%
13
+ [hops] 2024-09-23 16:14:19.118 | INFO | New best model: head accuracy 54.01% > 38.18%
14
+ [hops] 2024-09-23 16:14:29.492 | INFO | Epoch 3: train loss 1.4915 dev loss 1.2201 dev tag acc 55.22% dev head acc 62.64% dev deprel acc 72.20%
15
+ [hops] 2024-09-23 16:14:29.492 | INFO | New best model: head accuracy 62.64% > 54.01%
16
+ [hops] 2024-09-23 16:14:39.867 | INFO | Epoch 4: train loss 1.2642 dev loss 1.0422 dev tag acc 64.74% dev head acc 68.98% dev deprel acc 78.59%
17
+ [hops] 2024-09-23 16:14:39.868 | INFO | New best model: head accuracy 68.98% > 62.64%
18
+ [hops] 2024-09-23 16:14:50.176 | INFO | Epoch 5: train loss 1.0661 dev loss 0.8861 dev tag acc 73.28% dev head acc 72.64% dev deprel acc 82.30%
19
+ [hops] 2024-09-23 16:14:50.176 | INFO | New best model: head accuracy 72.64% > 68.98%
20
+ [hops] 2024-09-23 16:15:00.035 | INFO | Epoch 6: train loss 0.9110 dev loss 0.8110 dev tag acc 79.97% dev head acc 75.34% dev deprel acc 83.54%
21
+ [hops] 2024-09-23 16:15:00.036 | INFO | New best model: head accuracy 75.34% > 72.64%
22
+ [hops] 2024-09-23 16:15:10.484 | INFO | Epoch 7: train loss 0.7924 dev loss 0.7090 dev tag acc 82.43% dev head acc 77.50% dev deprel acc 85.19%
23
+ [hops] 2024-09-23 16:15:10.485 | INFO | New best model: head accuracy 77.50% > 75.34%
24
+ [hops] 2024-09-23 16:15:20.718 | INFO | Epoch 8: train loss 0.6980 dev loss 0.6438 dev tag acc 84.23% dev head acc 78.11% dev deprel acc 86.49%
25
+ [hops] 2024-09-23 16:15:20.719 | INFO | New best model: head accuracy 78.11% > 77.50%
26
+ [hops] 2024-09-23 16:15:30.573 | INFO | Epoch 9: train loss 0.6270 dev loss 0.6051 dev tag acc 86.31% dev head acc 80.42% dev deprel acc 86.95%
27
+ [hops] 2024-09-23 16:15:30.574 | INFO | New best model: head accuracy 80.42% > 78.11%
28
+ [hops] 2024-09-23 16:15:40.486 | INFO | Epoch 10: train loss 0.5629 dev loss 0.5624 dev tag acc 89.32% dev head acc 80.31% dev deprel acc 87.99%
29
+ [hops] 2024-09-23 16:15:48.515 | INFO | Epoch 11: train loss 0.5067 dev loss 0.5517 dev tag acc 90.65% dev head acc 81.64% dev deprel acc 88.35%
30
+ [hops] 2024-09-23 16:15:48.515 | INFO | New best model: head accuracy 81.64% > 80.42%
31
+ [hops] 2024-09-23 16:15:58.296 | INFO | Epoch 12: train loss 0.4651 dev loss 0.5284 dev tag acc 92.42% dev head acc 82.31% dev deprel acc 88.75%
32
+ [hops] 2024-09-23 16:15:58.297 | INFO | New best model: head accuracy 82.31% > 81.64%
33
+ [hops] 2024-09-23 16:16:08.157 | INFO | Epoch 13: train loss 0.4250 dev loss 0.5025 dev tag acc 93.53% dev head acc 83.33% dev deprel acc 89.65%
34
+ [hops] 2024-09-23 16:16:08.158 | INFO | New best model: head accuracy 83.33% > 82.31%
35
+ [hops] 2024-09-23 16:16:17.917 | INFO | Epoch 14: train loss 0.3938 dev loss 0.4943 dev tag acc 93.93% dev head acc 84.46% dev deprel acc 89.86%
36
+ [hops] 2024-09-23 16:16:17.918 | INFO | New best model: head accuracy 84.46% > 83.33%
37
+ [hops] 2024-09-23 16:16:27.703 | INFO | Epoch 15: train loss 0.3636 dev loss 0.4950 dev tag acc 94.81% dev head acc 84.71% dev deprel acc 89.90%
38
+ [hops] 2024-09-23 16:16:27.704 | INFO | New best model: head accuracy 84.71% > 84.46%
39
+ [hops] 2024-09-23 16:16:37.566 | INFO | Epoch 16: train loss 0.3358 dev loss 0.4867 dev tag acc 95.23% dev head acc 85.02% dev deprel acc 90.12%
40
+ [hops] 2024-09-23 16:16:37.567 | INFO | New best model: head accuracy 85.02% > 84.71%
41
+ [hops] 2024-09-23 16:16:47.274 | INFO | Epoch 17: train loss 0.3155 dev loss 0.4794 dev tag acc 95.56% dev head acc 85.76% dev deprel acc 90.02%
42
+ [hops] 2024-09-23 16:16:47.275 | INFO | New best model: head accuracy 85.76% > 85.02%
43
+ [hops] 2024-09-23 16:16:56.984 | INFO | Epoch 18: train loss 0.2961 dev loss 0.4793 dev tag acc 95.86% dev head acc 85.64% dev deprel acc 90.34%
44
+ [hops] 2024-09-23 16:17:04.954 | INFO | Epoch 19: train loss 0.2805 dev loss 0.4874 dev tag acc 96.20% dev head acc 86.08% dev deprel acc 90.62%
45
+ [hops] 2024-09-23 16:17:04.955 | INFO | New best model: head accuracy 86.08% > 85.76%
46
+ [hops] 2024-09-23 16:17:15.143 | INFO | Epoch 20: train loss 0.2617 dev loss 0.4873 dev tag acc 96.33% dev head acc 86.28% dev deprel acc 90.91%
47
+ [hops] 2024-09-23 16:17:15.145 | INFO | New best model: head accuracy 86.28% > 86.08%
48
+ [hops] 2024-09-23 16:17:25.286 | INFO | Epoch 21: train loss 0.2492 dev loss 0.4855 dev tag acc 96.44% dev head acc 86.26% dev deprel acc 91.19%
49
+ [hops] 2024-09-23 16:17:33.055 | INFO | Epoch 22: train loss 0.2353 dev loss 0.5254 dev tag acc 96.52% dev head acc 86.71% dev deprel acc 91.00%
50
+ [hops] 2024-09-23 16:17:33.056 | INFO | New best model: head accuracy 86.71% > 86.28%
51
+ [hops] 2024-09-23 16:17:42.549 | INFO | Epoch 23: train loss 0.2193 dev loss 0.5271 dev tag acc 96.61% dev head acc 86.32% dev deprel acc 91.24%
52
+ [hops] 2024-09-23 16:17:50.233 | INFO | Epoch 24: train loss 0.2127 dev loss 0.5132 dev tag acc 96.79% dev head acc 86.76% dev deprel acc 91.51%
53
+ [hops] 2024-09-23 16:17:50.234 | INFO | New best model: head accuracy 86.76% > 86.71%
54
+ [hops] 2024-09-23 16:18:00.614 | INFO | Epoch 25: train loss 0.2009 dev loss 0.5119 dev tag acc 96.57% dev head acc 86.68% dev deprel acc 91.51%
55
+ [hops] 2024-09-23 16:18:08.646 | INFO | Epoch 26: train loss 0.1921 dev loss 0.5223 dev tag acc 96.88% dev head acc 87.05% dev deprel acc 91.42%
56
+ [hops] 2024-09-23 16:18:08.647 | INFO | New best model: head accuracy 87.05% > 86.76%
57
+ [hops] 2024-09-23 16:18:18.112 | INFO | Epoch 27: train loss 0.1791 dev loss 0.5547 dev tag acc 97.13% dev head acc 86.99% dev deprel acc 91.39%
58
+ [hops] 2024-09-23 16:18:25.772 | INFO | Epoch 28: train loss 0.1744 dev loss 0.5114 dev tag acc 97.37% dev head acc 87.30% dev deprel acc 91.60%
59
+ [hops] 2024-09-23 16:18:25.773 | INFO | New best model: head accuracy 87.30% > 87.05%
60
+ [hops] 2024-09-23 16:18:35.209 | INFO | Epoch 29: train loss 0.1659 dev loss 0.5170 dev tag acc 97.31% dev head acc 87.43% dev deprel acc 91.83%
61
+ [hops] 2024-09-23 16:18:35.210 | INFO | New best model: head accuracy 87.43% > 87.30%
62
+ [hops] 2024-09-23 16:18:45.102 | INFO | Epoch 30: train loss 0.1572 dev loss 0.5548 dev tag acc 97.38% dev head acc 87.60% dev deprel acc 91.77%
63
+ [hops] 2024-09-23 16:18:45.102 | INFO | New best model: head accuracy 87.60% > 87.43%
64
+ [hops] 2024-09-23 16:18:55.060 | INFO | Epoch 31: train loss 0.1517 dev loss 0.5943 dev tag acc 97.39% dev head acc 87.63% dev deprel acc 91.78%
65
+ [hops] 2024-09-23 16:18:55.061 | INFO | New best model: head accuracy 87.63% > 87.60%
66
+ [hops] 2024-09-23 16:19:05.039 | INFO | Epoch 32: train loss 0.1458 dev loss 0.5661 dev tag acc 97.42% dev head acc 87.91% dev deprel acc 91.85%
67
+ [hops] 2024-09-23 16:19:05.040 | INFO | New best model: head accuracy 87.91% > 87.63%
68
+ [hops] 2024-09-23 16:19:14.549 | INFO | Epoch 33: train loss 0.1395 dev loss 0.5671 dev tag acc 97.52% dev head acc 87.96% dev deprel acc 91.90%
69
+ [hops] 2024-09-23 16:19:14.550 | INFO | New best model: head accuracy 87.96% > 87.91%
70
+ [hops] 2024-09-23 16:19:24.043 | INFO | Epoch 34: train loss 0.1338 dev loss 0.5757 dev tag acc 97.52% dev head acc 88.20% dev deprel acc 92.09%
71
+ [hops] 2024-09-23 16:19:24.044 | INFO | New best model: head accuracy 88.20% > 87.96%
72
+ [hops] 2024-09-23 16:19:33.500 | INFO | Epoch 35: train loss 0.1298 dev loss 0.5739 dev tag acc 97.57% dev head acc 88.23% dev deprel acc 92.11%
73
+ [hops] 2024-09-23 16:19:33.501 | INFO | New best model: head accuracy 88.23% > 88.20%
74
+ [hops] 2024-09-23 16:19:42.967 | INFO | Epoch 36: train loss 0.1229 dev loss 0.5932 dev tag acc 97.61% dev head acc 87.88% dev deprel acc 92.20%
75
+ [hops] 2024-09-23 16:19:50.940 | INFO | Epoch 37: train loss 0.1197 dev loss 0.6220 dev tag acc 97.62% dev head acc 88.34% dev deprel acc 91.96%
76
+ [hops] 2024-09-23 16:19:50.941 | INFO | New best model: head accuracy 88.34% > 88.23%
77
+ [hops] 2024-09-23 16:20:00.495 | INFO | Epoch 38: train loss 0.1157 dev loss 0.6152 dev tag acc 97.80% dev head acc 88.19% dev deprel acc 92.25%
78
+ [hops] 2024-09-23 16:20:09.048 | INFO | Epoch 39: train loss 0.1110 dev loss 0.6413 dev tag acc 97.75% dev head acc 88.12% dev deprel acc 92.16%
79
+ [hops] 2024-09-23 16:20:16.738 | INFO | Epoch 40: train loss 0.1085 dev loss 0.6398 dev tag acc 97.85% dev head acc 88.15% dev deprel acc 92.10%
80
+ [hops] 2024-09-23 16:20:25.222 | INFO | Epoch 41: train loss 0.1037 dev loss 0.6470 dev tag acc 97.89% dev head acc 88.26% dev deprel acc 92.18%
81
+ [hops] 2024-09-23 16:20:33.433 | INFO | Epoch 42: train loss 0.1017 dev loss 0.6571 dev tag acc 97.86% dev head acc 88.01% dev deprel acc 92.11%
82
+ [hops] 2024-09-23 16:20:41.541 | INFO | Epoch 43: train loss 0.1002 dev loss 0.6578 dev tag acc 97.91% dev head acc 88.16% dev deprel acc 92.24%
83
+ [hops] 2024-09-23 16:20:49.568 | INFO | Epoch 44: train loss 0.0947 dev loss 0.6615 dev tag acc 97.95% dev head acc 88.47% dev deprel acc 92.31%
84
+ [hops] 2024-09-23 16:20:49.569 | INFO | New best model: head accuracy 88.47% > 88.34%
85
+ [hops] 2024-09-23 16:20:59.008 | INFO | Epoch 45: train loss 0.0922 dev loss 0.6607 dev tag acc 97.95% dev head acc 88.04% dev deprel acc 92.31%
86
+ [hops] 2024-09-23 16:21:06.657 | INFO | Epoch 46: train loss 0.0900 dev loss 0.6693 dev tag acc 98.02% dev head acc 88.31% dev deprel acc 92.22%
87
+ [hops] 2024-09-23 16:21:14.318 | INFO | Epoch 47: train loss 0.0843 dev loss 0.6898 dev tag acc 97.93% dev head acc 88.45% dev deprel acc 92.34%
88
+ [hops] 2024-09-23 16:21:21.994 | INFO | Epoch 48: train loss 0.0823 dev loss 0.6972 dev tag acc 98.08% dev head acc 88.53% dev deprel acc 92.38%
89
+ [hops] 2024-09-23 16:21:21.995 | INFO | New best model: head accuracy 88.53% > 88.47%
90
+ [hops] 2024-09-23 16:21:31.473 | INFO | Epoch 49: train loss 0.0809 dev loss 0.6935 dev tag acc 98.13% dev head acc 88.57% dev deprel acc 92.51%
91
+ [hops] 2024-09-23 16:21:31.474 | INFO | New best model: head accuracy 88.57% > 88.53%
92
+ [hops] 2024-09-23 16:21:41.678 | INFO | Epoch 50: train loss 0.0764 dev loss 0.7254 dev tag acc 98.09% dev head acc 88.46% dev deprel acc 92.61%
93
+ [hops] 2024-09-23 16:21:49.318 | INFO | Epoch 51: train loss 0.0743 dev loss 0.7382 dev tag acc 98.05% dev head acc 88.35% dev deprel acc 92.48%
94
+ [hops] 2024-09-23 16:21:56.978 | INFO | Epoch 52: train loss 0.0765 dev loss 0.7135 dev tag acc 98.13% dev head acc 88.49% dev deprel acc 92.55%
95
+ [hops] 2024-09-23 16:22:04.687 | INFO | Epoch 53: train loss 0.0736 dev loss 0.7191 dev tag acc 98.10% dev head acc 88.55% dev deprel acc 92.50%
96
+ [hops] 2024-09-23 16:22:12.515 | INFO | Epoch 54: train loss 0.0703 dev loss 0.7270 dev tag acc 98.12% dev head acc 88.46% dev deprel acc 92.49%
97
+ [hops] 2024-09-23 16:22:20.205 | INFO | Epoch 55: train loss 0.0693 dev loss 0.7236 dev tag acc 98.13% dev head acc 88.58% dev deprel acc 92.31%
98
+ [hops] 2024-09-23 16:22:20.205 | INFO | New best model: head accuracy 88.58% > 88.57%
99
+ [hops] 2024-09-23 16:22:29.691 | INFO | Epoch 56: train loss 0.0676 dev loss 0.7135 dev tag acc 98.12% dev head acc 88.74% dev deprel acc 92.52%
100
+ [hops] 2024-09-23 16:22:29.691 | INFO | New best model: head accuracy 88.74% > 88.58%
101
+ [hops] 2024-09-23 16:22:39.168 | INFO | Epoch 57: train loss 0.0656 dev loss 0.7150 dev tag acc 98.18% dev head acc 88.74% dev deprel acc 92.42%
102
+ [hops] 2024-09-23 16:22:46.828 | INFO | Epoch 58: train loss 0.0653 dev loss 0.7252 dev tag acc 98.13% dev head acc 88.66% dev deprel acc 92.40%
103
+ [hops] 2024-09-23 16:22:54.486 | INFO | Epoch 59: train loss 0.0623 dev loss 0.7325 dev tag acc 98.13% dev head acc 88.74% dev deprel acc 92.43%
104
+ [hops] 2024-09-23 16:23:02.154 | INFO | Epoch 60: train loss 0.0619 dev loss 0.7384 dev tag acc 98.13% dev head acc 88.79% dev deprel acc 92.41%
105
+ [hops] 2024-09-23 16:23:02.155 | INFO | New best model: head accuracy 88.79% > 88.74%
106
+ [hops] 2024-09-23 16:23:11.676 | INFO | Epoch 61: train loss 0.0638 dev loss 0.7352 dev tag acc 98.15% dev head acc 88.80% dev deprel acc 92.47%
107
+ [hops] 2024-09-23 16:23:11.677 | INFO | New best model: head accuracy 88.80% > 88.79%
108
+ [hops] 2024-09-23 16:23:21.684 | INFO | Epoch 62: train loss 0.0625 dev loss 0.7389 dev tag acc 98.15% dev head acc 88.80% dev deprel acc 92.50%
109
+ [hops] 2024-09-23 16:23:29.540 | INFO | Epoch 63: train loss 0.0602 dev loss 0.7405 dev tag acc 98.16% dev head acc 88.82% dev deprel acc 92.50%
110
+ [hops] 2024-09-23 16:23:29.541 | INFO | New best model: head accuracy 88.82% > 88.80%
111
+ [hops] 2024-09-23 16:23:35.588 | WARNING | You're using a RobertaTokenizerFast tokenizer. Please note that with a fast tokenizer, using the `__call__` method is faster than using a method to encode the text followed by a call to the `pad` method to get a padded encoding.
112
+ [hops] 2024-09-23 16:23:41.549 | WARNING | You're using a RobertaTokenizerFast tokenizer. Please note that with a fast tokenizer, using the `__call__` method is faster than using a method to encode the text followed by a call to the `pad` method to get a padded encoding.
113
+ [hops] 2024-09-23 16:23:43.669 | INFO | Metrics for Rhapsodie-camembertv2_base_p2_17k_last_layer+rand_seed=123
114
+ ───────────────────────────────
115
+ Split UPOS UAS LAS
116
+ ───────────────────────────────
117
+ Dev 98.16 88.88 85.00
118
+ Test 97.56 88.98 84.50
119
+ ───────────────────────────────
120
+