Hugging Face's logo

almanach
/
camembertav2-base-sequoia

Token Classification
French
deberta-v2
Eval Results
Model card Files Community
wissamantoun commited on
Commit
2cf4cca
•
1 Parent(s): ecbc748

Upload folder using huggingface_hub

Browse files
Files changed (16) hide show
  1. README.md +161 -0
  2. camembertav2_base_p2_17k_last_layer.yaml +32 -0
  3. fr_sequoia-ud-dev.parsed.conllu +0 -0
  4. fr_sequoia-ud-test.parsed.conllu +0 -0
  5. model/config.json +1 -0
  6. model/lexers/camembertav2_base_p2_17k_last_layer/config.json +1 -0
  7. model/lexers/camembertav2_base_p2_17k_last_layer/model/config.json +41 -0
  8. model/lexers/camembertav2_base_p2_17k_last_layer/model/special_tokens_map.json +51 -0
  9. model/lexers/camembertav2_base_p2_17k_last_layer/model/tokenizer.json +0 -0
  10. model/lexers/camembertav2_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 +0 -0
  15. model/weights.pt +3 -0
  16. train.log +111 -0
README.md ADDED
@@ -0,0 +1,161 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language: fr
3
+ license: mit
4
+ tags:
5
+ - deberta-v2
6
+ - token-classification
7
+ base_model: almanach/camembertav2-base
8
+ datasets:
9
+ - Sequoia
10
+ metrics:
11
+ - las
12
+ - upos
13
+ model-index:
14
+ - name: almanach/camembertav2-base-sequoia
15
+ results:
16
+ - task:
17
+ type: token-classification
18
+ name: Part-of-Speech Tagging
19
+ dataset:
20
+ type: Sequoia
21
+ name: Sequoia
22
+ metrics:
23
+ - name: upos
24
+ type: upos
25
+ value: 0.99423
26
+ verified: false
27
+ - task:
28
+ type: token-classification
29
+ name: Dependency Parsing
30
+ dataset:
31
+ type: Sequoia
32
+ name: Sequoia
33
+ metrics:
34
+ - name: las
35
+ type: las
36
+ value: 0.94883
37
+ verified: false
38
+ ---
39
+
40
+ # Model Card for almanach/camembertav2-base-sequoia
41
+
42
+ almanach/camembertav2-base-sequoia is a deberta-v2 model for token classification. It is trained on the Sequoia dataset for the task of Part-of-Speech Tagging and Dependency Parsing.
43
+ The model achieves an f1 score of on the Sequoia dataset.
44
+
45
+ The model is part of the almanach/camembertav2-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:** deberta-v2
53
+ - **Language(s) (NLP):** French
54
+ - **License:** MIT
55
+ - **Finetuned from model :** almanach/camembertav2-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 Sequoia 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: camembertav2_base_p2_17k_last_layer
106
+ type: bert
107
+ model: /scratch/camembertv2/runs/models/camembertav2-base-bf16/post/ckpt-p2-17000/pt/discriminator/
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:** 0.99423
126
+ **LAS:** 0.94883
127
+
128
+ ## Technical Specifications
129
+
130
+ ### Model Architecture and Objective
131
+
132
+ deberta-v2 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
+ ```
camembertav2_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: camembertav2_base_p2_17k_last_layer
19
+ type: bert
20
+ model: /scratch/camembertv2/runs/models/camembertav2-base-bf16/post/ckpt-p2-17000/pt/discriminator/
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_sequoia-ud-dev.parsed.conllu ADDED
The diff for this file is too large to render. See raw diff
 
fr_sequoia-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:caus", "aux:pass", "aux:tense", "case", "cc", "ccomp", "conj", "cop", "csubj", "csubj:pass", "dep", "det", "discourse", "dislocated", "expl:comp", "expl:pass", "expl:subj", "fixed", "flat:foreign", "flat:name", "goeswith", "iobj", "iobj:agent", "mark", "nmod", "nsubj", "nsubj:caus", "nsubj:pass", "nummod", "obj", "obj:agent", "obl:agent", "obl:arg", "obl:mod", "orphan", "parataxis", "punct", "root", "vocative", "xcomp"], "mlp_dropout": 0.5, "tagset": ["ADJ", "ADP", "ADV", "AUX", "CCONJ", "DET", "INTJ", "NOUN", "NUM", "PRON", "PROPN", "PUNCT", "SCONJ", "SYM", "VERB", "X"], "lexers": {"word_embeddings": "words", "char_level_embeddings": "chars_rnn", "fasttext": "fasttext", "camembertav2_base_p2_17k_last_layer": "bert"}, "multitask_loss": "sum"}
model/lexers/camembertav2_base_p2_17k_last_layer/config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"layers": [11], "subwords_reduction": "mean", "weight_layers": false}
model/lexers/camembertav2_base_p2_17k_last_layer/model/config.json ADDED
@@ -0,0 +1,41 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "/scratch/camembertv2/runs/models/camembertav2-base-bf16/post/ckpt-p2-17000/pt/discriminator/",
3
+ "architectures": [
4
+ "DebertaV2Model"
5
+ ],
6
+ "attention_probs_dropout_prob": 0.1,
7
+ "bos_token_id": 1,
8
+ "conv_act": "gelu",
9
+ "conv_kernel_size": 0,
10
+ "embedding_size": 768,
11
+ "eos_token_id": 2,
12
+ "hidden_act": "gelu",
13
+ "hidden_dropout_prob": 0.1,
14
+ "hidden_size": 768,
15
+ "initializer_range": 0.02,
16
+ "intermediate_size": 3072,
17
+ "layer_norm_eps": 1e-07,
18
+ "max_position_embeddings": 1024,
19
+ "max_relative_positions": -1,
20
+ "model_name": "camembertav2-base-bf16",
21
+ "model_type": "deberta-v2",
22
+ "norm_rel_ebd": "layer_norm",
23
+ "num_attention_heads": 12,
24
+ "num_hidden_layers": 12,
25
+ "pad_token_id": 0,
26
+ "pooler_dropout": 0,
27
+ "pooler_hidden_act": "gelu",
28
+ "pooler_hidden_size": 768,
29
+ "pos_att_type": [
30
+ "p2c",
31
+ "c2p"
32
+ ],
33
+ "position_biased_input": false,
34
+ "position_buckets": 256,
35
+ "relative_attention": true,
36
+ "share_att_key": true,
37
+ "torch_dtype": "float32",
38
+ "transformers_version": "4.44.2",
39
+ "type_vocab_size": 0,
40
+ "vocab_size": 32768
41
+ }
model/lexers/camembertav2_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/camembertav2_base_p2_17k_last_layer/model/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
model/lexers/camembertav2_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>", " ", "!", "\"", "$", "%", "&", "'", "(", ")", "+", ",", "-", ".", "/", "0", "1", "2", "3", "4", "5", "6", "7", "8", "9", ":", ";", "<", "=", "?", "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", "\u00a9", "\u00b0", "\u00b1", "\u00bd", "\u00c0", "\u00c9", "\u00ce", "\u00df", "\u00e0", "\u00e1", "\u00e2", "\u00e4", "\u00e7", "\u00e8", "\u00e9", "\u00ea", "\u00eb", "\u00ee", "\u00ef", "\u00f3", "\u00f4", "\u00f6", "\u00f9", "\u00fb"]}
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:cf09011cf6593888c882b0464e1b82ae5a3d05fce8e5c2861014f45557861568
3
+ size 801050258
model/lexers/word_embeddings/config.json ADDED
The diff for this file is too large to render. See raw diff
 
model/weights.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6626b13e315a9bdd4c63e4321755366a968ab2595f4cb625cf45b67312ba5790
3
+ size 1745757420
train.log ADDED
@@ -0,0 +1,111 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [hops] 2024-09-24 16:01:20.681 | INFO | Initializing a parser from /workspace/configs/exp_camembertv2/camembertav2_base_p2_17k_last_layer.yaml
2
+ [hops] 2024-09-24 16:01:20.730 | INFO | Generating a FastText model from the treebank
3
+ [hops] 2024-09-24 16:01:20.745 | INFO | Training fasttext model
4
+ [hops] 2024-09-24 16:01:28.399 | INFO | Start training on cuda:0
5
+ [hops] 2024-09-24 16:01:28.403 | 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.
6
+ [hops] 2024-09-24 16:01:47.740 | INFO | Epoch 0: train loss 2.7794 dev loss 1.9428 dev tag acc 49.62% dev head acc 27.95% dev deprel acc 53.97%
7
+ [hops] 2024-09-24 16:01:47.741 | INFO | New best model: head accuracy 27.95% > 0.00%
8
+ [hops] 2024-09-24 16:02:09.374 | INFO | Epoch 1: train loss 1.5546 dev loss 1.0005 dev tag acc 72.25% dev head acc 61.63% dev deprel acc 80.28%
9
+ [hops] 2024-09-24 16:02:09.375 | INFO | New best model: head accuracy 61.63% > 27.95%
10
+ [hops] 2024-09-24 16:02:31.031 | INFO | Epoch 2: train loss 0.9383 dev loss 0.6218 dev tag acc 80.12% dev head acc 79.32% dev deprel acc 87.02%
11
+ [hops] 2024-09-24 16:02:31.032 | INFO | New best model: head accuracy 79.32% > 61.63%
12
+ [hops] 2024-09-24 16:02:51.968 | INFO | Epoch 3: train loss 0.6428 dev loss 0.4519 dev tag acc 87.60% dev head acc 84.25% dev deprel acc 90.09%
13
+ [hops] 2024-09-24 16:02:51.969 | INFO | New best model: head accuracy 84.25% > 79.32%
14
+ [hops] 2024-09-24 16:03:13.587 | INFO | Epoch 4: train loss 0.4912 dev loss 0.3727 dev tag acc 91.58% dev head acc 85.25% dev deprel acc 92.18%
15
+ [hops] 2024-09-24 16:03:13.588 | INFO | New best model: head accuracy 85.25% > 84.25%
16
+ [hops] 2024-09-24 16:03:34.880 | INFO | Epoch 5: train loss 0.3884 dev loss 0.3005 dev tag acc 95.05% dev head acc 88.23% dev deprel acc 93.64%
17
+ [hops] 2024-09-24 16:03:34.881 | INFO | New best model: head accuracy 88.23% > 85.25%
18
+ [hops] 2024-09-24 16:03:56.210 | INFO | Epoch 6: train loss 0.3135 dev loss 0.2582 dev tag acc 96.33% dev head acc 90.19% dev deprel acc 94.54%
19
+ [hops] 2024-09-24 16:03:56.211 | INFO | New best model: head accuracy 90.19% > 88.23%
20
+ [hops] 2024-09-24 16:04:18.250 | INFO | Epoch 7: train loss 0.2602 dev loss 0.2364 dev tag acc 96.99% dev head acc 90.92% dev deprel acc 95.35%
21
+ [hops] 2024-09-24 16:04:18.251 | INFO | New best model: head accuracy 90.92% > 90.19%
22
+ [hops] 2024-09-24 16:04:39.889 | INFO | Epoch 8: train loss 0.2206 dev loss 0.2207 dev tag acc 97.62% dev head acc 91.55% dev deprel acc 95.82%
23
+ [hops] 2024-09-24 16:04:39.890 | INFO | New best model: head accuracy 91.55% > 90.92%
24
+ [hops] 2024-09-24 16:05:00.941 | INFO | Epoch 9: train loss 0.1885 dev loss 0.2092 dev tag acc 97.86% dev head acc 92.34% dev deprel acc 96.26%
25
+ [hops] 2024-09-24 16:05:00.942 | INFO | New best model: head accuracy 92.34% > 91.55%
26
+ [hops] 2024-09-24 16:05:22.018 | INFO | Epoch 10: train loss 0.1633 dev loss 0.1818 dev tag acc 98.23% dev head acc 92.87% dev deprel acc 96.84%
27
+ [hops] 2024-09-24 16:05:22.019 | INFO | New best model: head accuracy 92.87% > 92.34%
28
+ [hops] 2024-09-24 16:05:42.839 | INFO | Epoch 11: train loss 0.1444 dev loss 0.1800 dev tag acc 98.40% dev head acc 93.47% dev deprel acc 96.77%
29
+ [hops] 2024-09-24 16:05:42.840 | INFO | New best model: head accuracy 93.47% > 92.87%
30
+ [hops] 2024-09-24 16:06:04.127 | INFO | Epoch 12: train loss 0.1289 dev loss 0.1718 dev tag acc 98.58% dev head acc 93.67% dev deprel acc 97.08%
31
+ [hops] 2024-09-24 16:06:04.128 | INFO | New best model: head accuracy 93.67% > 93.47%
32
+ [hops] 2024-09-24 16:06:26.120 | INFO | Epoch 13: train loss 0.1136 dev loss 0.1875 dev tag acc 98.59% dev head acc 93.56% dev deprel acc 97.00%
33
+ [hops] 2024-09-24 16:06:45.617 | INFO | Epoch 14: train loss 0.1033 dev loss 0.1923 dev tag acc 98.85% dev head acc 93.73% dev deprel acc 97.03%
34
+ [hops] 2024-09-24 16:06:45.618 | INFO | New best model: head accuracy 93.73% > 93.67%
35
+ [hops] 2024-09-24 16:07:07.243 | INFO | Epoch 15: train loss 0.0938 dev loss 0.1859 dev tag acc 98.89% dev head acc 94.23% dev deprel acc 97.14%
36
+ [hops] 2024-09-24 16:07:07.244 | INFO | New best model: head accuracy 94.23% > 93.73%
37
+ [hops] 2024-09-24 16:07:29.285 | INFO | Epoch 16: train loss 0.0855 dev loss 0.1829 dev tag acc 98.89% dev head acc 94.30% dev deprel acc 97.31%
38
+ [hops] 2024-09-24 16:07:29.287 | INFO | New best model: head accuracy 94.30% > 94.23%
39
+ [hops] 2024-09-24 16:07:50.544 | INFO | Epoch 17: train loss 0.0789 dev loss 0.1872 dev tag acc 98.91% dev head acc 94.57% dev deprel acc 97.40%
40
+ [hops] 2024-09-24 16:07:50.545 | INFO | New best model: head accuracy 94.57% > 94.30%
41
+ [hops] 2024-09-24 16:08:12.269 | INFO | Epoch 18: train loss 0.0730 dev loss 0.1901 dev tag acc 98.95% dev head acc 94.59% dev deprel acc 97.29%
42
+ [hops] 2024-09-24 16:08:12.270 | INFO | New best model: head accuracy 94.59% > 94.57%
43
+ [hops] 2024-09-24 16:08:33.803 | INFO | Epoch 19: train loss 0.0665 dev loss 0.1852 dev tag acc 98.91% dev head acc 94.72% dev deprel acc 97.43%
44
+ [hops] 2024-09-24 16:08:33.804 | INFO | New best model: head accuracy 94.72% > 94.59%
45
+ [hops] 2024-09-24 16:08:55.220 | INFO | Epoch 20: train loss 0.0620 dev loss 0.2002 dev tag acc 98.96% dev head acc 95.00% dev deprel acc 97.45%
46
+ [hops] 2024-09-24 16:08:55.221 | INFO | New best model: head accuracy 95.00% > 94.72%
47
+ [hops] 2024-09-24 16:09:16.939 | INFO | Epoch 21: train loss 0.0574 dev loss 0.2063 dev tag acc 99.04% dev head acc 94.79% dev deprel acc 97.56%
48
+ [hops] 2024-09-24 16:09:36.037 | INFO | Epoch 22: train loss 0.0534 dev loss 0.2002 dev tag acc 99.06% dev head acc 95.04% dev deprel acc 97.55%
49
+ [hops] 2024-09-24 16:09:36.038 | INFO | New best model: head accuracy 95.04% > 95.00%
50
+ [hops] 2024-09-24 16:09:57.103 | INFO | Epoch 23: train loss 0.0503 dev loss 0.2077 dev tag acc 99.06% dev head acc 95.07% dev deprel acc 97.55%
51
+ [hops] 2024-09-24 16:09:57.103 | INFO | New best model: head accuracy 95.07% > 95.04%
52
+ [hops] 2024-09-24 16:10:18.368 | INFO | Epoch 24: train loss 0.0469 dev loss 0.2004 dev tag acc 99.06% dev head acc 95.43% dev deprel acc 97.68%
53
+ [hops] 2024-09-24 16:10:18.369 | INFO | New best model: head accuracy 95.43% > 95.07%
54
+ [hops] 2024-09-24 16:10:40.222 | INFO | Epoch 25: train loss 0.0432 dev loss 0.2043 dev tag acc 99.02% dev head acc 95.30% dev deprel acc 97.67%
55
+ [hops] 2024-09-24 16:10:59.739 | INFO | Epoch 26: train loss 0.0416 dev loss 0.2225 dev tag acc 99.06% dev head acc 95.08% dev deprel acc 97.48%
56
+ [hops] 2024-09-24 16:11:18.934 | INFO | Epoch 27: train loss 0.0375 dev loss 0.2118 dev tag acc 99.08% dev head acc 95.39% dev deprel acc 97.62%
57
+ [hops] 2024-09-24 16:11:37.455 | INFO | Epoch 28: train loss 0.0369 dev loss 0.2139 dev tag acc 99.08% dev head acc 95.28% dev deprel acc 97.76%
58
+ [hops] 2024-09-24 16:11:57.066 | INFO | Epoch 29: train loss 0.0351 dev loss 0.2086 dev tag acc 99.12% dev head acc 95.46% dev deprel acc 97.72%
59
+ [hops] 2024-09-24 16:11:57.067 | INFO | New best model: head accuracy 95.46% > 95.43%
60
+ [hops] 2024-09-24 16:12:18.271 | INFO | Epoch 30: train loss 0.0329 dev loss 0.2321 dev tag acc 99.06% dev head acc 95.49% dev deprel acc 97.58%
61
+ [hops] 2024-09-24 16:12:18.272 | INFO | New best model: head accuracy 95.49% > 95.46%
62
+ [hops] 2024-09-24 16:12:40.183 | INFO | Epoch 31: train loss 0.0320 dev loss 0.2237 dev tag acc 99.14% dev head acc 95.79% dev deprel acc 97.80%
63
+ [hops] 2024-09-24 16:12:40.184 | INFO | New best model: head accuracy 95.79% > 95.49%
64
+ [hops] 2024-09-24 16:13:01.544 | INFO | Epoch 32: train loss 0.0291 dev loss 0.2373 dev tag acc 99.10% dev head acc 95.38% dev deprel acc 97.79%
65
+ [hops] 2024-09-24 16:13:20.203 | INFO | Epoch 33: train loss 0.0288 dev loss 0.2393 dev tag acc 99.13% dev head acc 95.61% dev deprel acc 97.77%
66
+ [hops] 2024-09-24 16:13:38.371 | INFO | Epoch 34: train loss 0.0261 dev loss 0.2499 dev tag acc 99.08% dev head acc 95.62% dev deprel acc 97.76%
67
+ [hops] 2024-09-24 16:13:57.799 | INFO | Epoch 35: train loss 0.0254 dev loss 0.2435 dev tag acc 99.11% dev head acc 95.74% dev deprel acc 97.81%
68
+ [hops] 2024-09-24 16:14:15.970 | INFO | Epoch 36: train loss 0.0235 dev loss 0.2542 dev tag acc 99.15% dev head acc 95.53% dev deprel acc 97.85%
69
+ [hops] 2024-09-24 16:14:34.739 | INFO | Epoch 37: train loss 0.0226 dev loss 0.2540 dev tag acc 99.07% dev head acc 95.39% dev deprel acc 97.71%
70
+ [hops] 2024-09-24 16:14:53.771 | INFO | Epoch 38: train loss 0.0218 dev loss 0.2529 dev tag acc 99.09% dev head acc 95.24% dev deprel acc 97.74%
71
+ [hops] 2024-09-24 16:15:13.379 | INFO | Epoch 39: train loss 0.0206 dev loss 0.2571 dev tag acc 99.09% dev head acc 95.57% dev deprel acc 97.83%
72
+ [hops] 2024-09-24 16:15:32.562 | INFO | Epoch 40: train loss 0.0195 dev loss 0.2649 dev tag acc 99.19% dev head acc 95.64% dev deprel acc 97.80%
73
+ [hops] 2024-09-24 16:15:50.944 | INFO | Epoch 41: train loss 0.0194 dev loss 0.2632 dev tag acc 99.14% dev head acc 95.73% dev deprel acc 97.77%
74
+ [hops] 2024-09-24 16:16:09.974 | INFO | Epoch 42: train loss 0.0178 dev loss 0.2683 dev tag acc 99.21% dev head acc 95.81% dev deprel acc 97.85%
75
+ [hops] 2024-09-24 16:16:09.975 | INFO | New best model: head accuracy 95.81% > 95.79%
76
+ [hops] 2024-09-24 16:16:31.272 | INFO | Epoch 43: train loss 0.0162 dev loss 0.2753 dev tag acc 99.20% dev head acc 95.74% dev deprel acc 97.80%
77
+ [hops] 2024-09-24 16:16:49.960 | INFO | Epoch 44: train loss 0.0233 dev loss 0.2764 dev tag acc 99.21% dev head acc 95.75% dev deprel acc 97.84%
78
+ [hops] 2024-09-24 16:17:09.618 | INFO | Epoch 45: train loss 0.0157 dev loss 0.2860 dev tag acc 99.23% dev head acc 95.84% dev deprel acc 97.92%
79
+ [hops] 2024-09-24 16:17:09.619 | INFO | New best model: head accuracy 95.84% > 95.81%
80
+ [hops] 2024-09-24 16:17:31.311 | INFO | Epoch 46: train loss 0.0141 dev loss 0.2782 dev tag acc 99.24% dev head acc 95.67% dev deprel acc 97.92%
81
+ [hops] 2024-09-24 16:17:50.429 | INFO | Epoch 47: train loss 0.0135 dev loss 0.2823 dev tag acc 99.20% dev head acc 95.93% dev deprel acc 97.84%
82
+ [hops] 2024-09-24 16:17:50.430 | INFO | New best model: head accuracy 95.93% > 95.84%
83
+ [hops] 2024-09-24 16:18:12.154 | INFO | Epoch 48: train loss 0.0139 dev loss 0.2830 dev tag acc 99.17% dev head acc 95.77% dev deprel acc 97.81%
84
+ [hops] 2024-09-24 16:18:30.927 | INFO | Epoch 49: train loss 0.0129 dev loss 0.2882 dev tag acc 99.17% dev head acc 95.87% dev deprel acc 97.78%
85
+ [hops] 2024-09-24 16:18:50.677 | INFO | Epoch 50: train loss 0.0120 dev loss 0.2876 dev tag acc 99.18% dev head acc 95.82% dev deprel acc 97.80%
86
+ [hops] 2024-09-24 16:19:09.749 | INFO | Epoch 51: train loss 0.0115 dev loss 0.2998 dev tag acc 99.19% dev head acc 95.69% dev deprel acc 97.85%
87
+ [hops] 2024-09-24 16:19:28.709 | INFO | Epoch 52: train loss 0.0116 dev loss 0.2948 dev tag acc 99.22% dev head acc 95.76% dev deprel acc 97.95%
88
+ [hops] 2024-09-24 16:19:47.697 | INFO | Epoch 53: train loss 0.0102 dev loss 0.3000 dev tag acc 99.22% dev head acc 95.95% dev deprel acc 97.90%
89
+ [hops] 2024-09-24 16:19:47.698 | INFO | New best model: head accuracy 95.95% > 95.93%
90
+ [hops] 2024-09-24 16:20:09.017 | INFO | Epoch 54: train loss 0.0104 dev loss 0.3013 dev tag acc 99.23% dev head acc 96.02% dev deprel acc 97.89%
91
+ [hops] 2024-09-24 16:20:09.018 | INFO | New best model: head accuracy 96.02% > 95.95%
92
+ [hops] 2024-09-24 16:20:30.175 | INFO | Epoch 55: train loss 0.0105 dev loss 0.2964 dev tag acc 99.26% dev head acc 96.01% dev deprel acc 97.95%
93
+ [hops] 2024-09-24 16:20:50.301 | INFO | Epoch 56: train loss 0.0098 dev loss 0.2959 dev tag acc 99.23% dev head acc 95.97% dev deprel acc 97.89%
94
+ [hops] 2024-09-24 16:21:09.359 | INFO | Epoch 57: train loss 0.0091 dev loss 0.3062 dev tag acc 99.25% dev head acc 95.88% dev deprel acc 97.87%
95
+ [hops] 2024-09-24 16:21:28.289 | INFO | Epoch 58: train loss 0.0089 dev loss 0.3102 dev tag acc 99.25% dev head acc 95.87% dev deprel acc 97.85%
96
+ [hops] 2024-09-24 16:21:48.019 | INFO | Epoch 59: train loss 0.0089 dev loss 0.3086 dev tag acc 99.24% dev head acc 95.97% dev deprel acc 97.91%
97
+ [hops] 2024-09-24 16:22:06.944 | INFO | Epoch 60: train loss 0.0086 dev loss 0.3091 dev tag acc 99.25% dev head acc 95.98% dev deprel acc 97.91%
98
+ [hops] 2024-09-24 16:22:25.872 | INFO | Epoch 61: train loss 0.0080 dev loss 0.3121 dev tag acc 99.27% dev head acc 95.96% dev deprel acc 97.91%
99
+ [hops] 2024-09-24 16:22:44.686 | INFO | Epoch 62: train loss 0.0087 dev loss 0.3127 dev tag acc 99.25% dev head acc 96.02% dev deprel acc 97.91%
100
+ [hops] 2024-09-24 16:23:03.731 | INFO | Epoch 63: train loss 0.0083 dev loss 0.3123 dev tag acc 99.25% dev head acc 96.03% dev deprel acc 97.92%
101
+ [hops] 2024-09-24 16:23:03.732 | INFO | New best model: head accuracy 96.03% > 96.02%
102
+ [hops] 2024-09-24 16:23:11.038 | 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.
103
+ [hops] 2024-09-24 16:23:16.913 | 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.
104
+ [hops] 2024-09-24 16:23:18.926 | INFO | Metrics for Sequoia-camembertav2_base_p2_17k_last_layer+rand_seed=42
105
+ ───────────────────────────────
106
+ Split UPOS UAS LAS
107
+ ───────────────────────────────
108
+ Dev 99.25 96.08 94.89
109
+ Test 99.42 95.98 94.88
110
+ ───────────────────────────────
111
+