arthrod commited on
Commit
b5de386
·
verified ·
1 Parent(s): e7b7d8a

Training in progress, step 500

Browse files
.gitattributes CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ tokenizer.json filter=lfs diff=lfs merge=lfs -text
config.yaml ADDED
@@ -0,0 +1,117 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ run:
2
+ name: b-gliner-pii-token-multipii-v1-spam-v1
3
+ description: Domain adaptation for PII-in-spam using token-level GLiNER starting
4
+ from gliner_multi_pii-v1.
5
+ tags:
6
+ - pii
7
+ - spam
8
+ - gliner
9
+ - token_level
10
+ - domain-adapt
11
+ - multipii-v1
12
+ seed: 42
13
+ model:
14
+ model_name: microsoft/deberta-v3-base
15
+ name: gliner-pii-token-multipii-v1-spam
16
+ labels_encoder: null
17
+ labels_decoder: null
18
+ decoder_mode: null
19
+ full_decoder_context: true
20
+ blank_entity_prob: 0.1
21
+ decoder_loss_coef: 0.5
22
+ relations_layer: null
23
+ triples_layer: null
24
+ embed_rel_token: true
25
+ rel_token_index: -1
26
+ rel_token: <<REL>>
27
+ adjacency_loss_coef: 1.0
28
+ relation_loss_coef: 1.0
29
+ span_mode: token_level
30
+ max_width: 12
31
+ represent_spans: false
32
+ neg_spans_ratio: 1.0
33
+ hidden_size: 768
34
+ dropout: 0.25
35
+ fine_tune: true
36
+ subtoken_pooling: first
37
+ fuse_layers: false
38
+ post_fusion_schema: ''
39
+ num_post_fusion_layers: 1
40
+ num_rnn_layers: 0
41
+ max_len: 384
42
+ max_types: 100
43
+ max_neg_type_ratio: 2
44
+ words_splitter_type: whitespace
45
+ embed_ent_token: true
46
+ class_token_index: -1
47
+ ent_token: <<ENT>>
48
+ sep_token: <<SEP>>
49
+ vocab_size: -1
50
+ token_loss_coef: 1.0
51
+ span_loss_coef: 1.0
52
+ encoder_config: null
53
+ _attn_implementation: flash_attention_2
54
+ labels_encoder_config: null
55
+ labels_decoder_config: null
56
+ data:
57
+ root_dir: gliner_logs/gliner-pii-token-multipii-v1-spam-v1
58
+ train_data: arthrod/gliner-flex-pii-ready-v5::train
59
+ val_data_dir: arthrod/gliner-flex-pii-ready-v5::eval
60
+ training:
61
+ prev_path: urchade/gliner_multi_pii-v1
62
+ num_steps: 12000
63
+ scheduler_type: cosine
64
+ warmup_ratio: 0.05
65
+ train_batch_size: 32
66
+ eval_batch_size: 64
67
+ gradient_accumulation_steps: 1
68
+ max_grad_norm: 1.0
69
+ optimizer: adamw_torch
70
+ lr_encoder: 8.0e-06
71
+ lr_others: 4.0e-05
72
+ weight_decay_encoder: 0.01
73
+ weight_decay_other: 0.01
74
+ loss_alpha: 0.75
75
+ loss_gamma: 2.0
76
+ loss_prob_margin: 0.0
77
+ label_smoothing: 0.0
78
+ loss_reduction: sum
79
+ negatives: 2.0
80
+ masking: none
81
+ eval_every: 500
82
+ save_total_limit: 20
83
+ logging_steps: 10
84
+ bf16: true
85
+ fp16: false
86
+ use_cpu: false
87
+ dataloader_num_workers: 18
88
+ dataloader_pin_memory: true
89
+ dataloader_persistent_workers: false
90
+ dataloader_prefetch_factor: 2
91
+ freeze_components: null
92
+ compile_model: false
93
+ gradient_checkpointing: false
94
+ eval_steps: null
95
+ eval_on_start: false
96
+ lora:
97
+ enabled: false
98
+ r: 16
99
+ lora_alpha: 32
100
+ lora_dropout: 0.05
101
+ bias: none
102
+ target_modules:
103
+ - all-linear
104
+ task_type: TOKEN_CLS
105
+ modules_to_save: null
106
+ init_lora_weights: true
107
+ use_rslora: false
108
+ fan_in_fan_out: false
109
+ environment:
110
+ push_to_hub: true
111
+ hub_model_id: arthrod/gliner-pii-token-multipii-v1-spam-v1
112
+ hf_token: null
113
+ report_to: wandb
114
+ wandb_project: gliner-pii
115
+ wandb_entity: null
116
+ wandb_api_key: null
117
+ cuda_visible_devices: '0'
pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2de5a72ba0f93e12dd7caf7c3826d37ebc3f99138715338bff92da18e87f46c5
3
+ size 1155880867
summary_20260224T084509Z.txt ADDED
@@ -0,0 +1,118 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ================================================================================
2
+ Configuration Validation Summary
3
+ ================================================================================
4
+ run.name OK b-gliner-pii-token-multipii-v1-spam-v1
5
+ run.description OK Domain adaptation for PII-in-spam using token-level GLiNER starting from gliner_multi_pii-v1.
6
+ run.tags OK ['pii', 'spam', 'gliner', 'token_level', 'domain-adapt', 'multipii-v1']
7
+ run.seed OK 42
8
+ model.model_name OK microsoft/deberta-v3-base
9
+ model.name OK gliner-pii-token-multipii-v1-spam
10
+ model.labels_encoder OK None
11
+ model.labels_decoder OK None
12
+ model.decoder_mode OK None
13
+ model.full_decoder_context OK True
14
+ model.blank_entity_prob OK 0.1
15
+ model.decoder_loss_coef OK 0.5
16
+ model.relations_layer OK None
17
+ model.triples_layer OK None
18
+ model.embed_rel_token OK True
19
+ model.rel_token_index OK -1
20
+ model.rel_token OK <<REL>>
21
+ model.adjacency_loss_coef OK 1.0
22
+ model.relation_loss_coef OK 1.0
23
+ model.span_mode OK token_level
24
+ model.max_width OK 12
25
+ model.represent_spans OK False
26
+ model.neg_spans_ratio OK 1.0
27
+ model.hidden_size OK 768
28
+ model.dropout OK 0.25
29
+ model.fine_tune OK True
30
+ model.subtoken_pooling OK first
31
+ model.fuse_layers OK False
32
+ model.post_fusion_schema OK
33
+ model.num_post_fusion_layers OK 1
34
+ model.num_rnn_layers OK 0
35
+ model.max_len OK 384
36
+ model.max_types OK 100
37
+ model.max_neg_type_ratio OK 2
38
+ model.words_splitter_type OK whitespace
39
+ model.embed_ent_token OK True
40
+ model.class_token_index OK -1
41
+ model.ent_token OK <<ENT>>
42
+ model.sep_token OK <<SEP>>
43
+ model.token_loss_coef OK 1.0
44
+ model.span_loss_coef OK 1.0
45
+ model.encoder_config OK None
46
+ model.labels_encoder_config DEFAULT None
47
+ model.labels_decoder_config DEFAULT None
48
+ model._attn_implementation OK flash_attention_2
49
+ model.vocab_size OK -1
50
+ data.root_dir OK gliner_logs/gliner-pii-token-multipii-v1-spam-v1
51
+ data.train_data OK arthrod/gliner-flex-pii-ready-v5::train
52
+ data.val_data_dir OK arthrod/gliner-flex-pii-ready-v5::eval
53
+ training.prev_path OK urchade/gliner_multi_pii-v1
54
+ training.num_steps OK 12000
55
+ training.scheduler_type OK cosine
56
+ training.warmup_ratio OK 0.05
57
+ training.train_batch_size OK 32
58
+ training.eval_batch_size OK 64
59
+ training.gradient_accumulation_steps OK 1
60
+ training.max_grad_norm OK 1.0
61
+ training.optimizer OK adamw_torch
62
+ training.lr_encoder OK 8e-06
63
+ training.lr_others OK 4e-05
64
+ training.weight_decay_encoder OK 0.01
65
+ training.weight_decay_other OK 0.01
66
+ training.loss_alpha OK 0.75
67
+ training.loss_gamma OK 2.0
68
+ training.loss_prob_margin OK 0.0
69
+ training.label_smoothing OK 0.0
70
+ training.loss_reduction OK sum
71
+ training.negatives OK 2.0
72
+ training.masking OK none
73
+ training.eval_every OK 500
74
+ training.save_total_limit OK 20
75
+ training.logging_steps OK 10
76
+ training.bf16 OK True
77
+ training.fp16 OK False
78
+ training.use_cpu OK False
79
+ training.dataloader_num_workers OK 18
80
+ training.dataloader_pin_memory OK True
81
+ training.dataloader_persistent_workers OK False
82
+ training.dataloader_prefetch_factor OK 2
83
+ training.freeze_components OK None
84
+ training.gradient_checkpointing DEFAULT False
85
+ training.eval_steps DEFAULT None
86
+ training.eval_on_start DEFAULT False
87
+ training.compile_model OK False
88
+ lora.enabled OK False
89
+ lora.r OK 16
90
+ lora.lora_alpha OK 32
91
+ lora.lora_dropout OK 0.05
92
+ lora.bias OK none
93
+ lora.target_modules OK ['all-linear']
94
+ lora.task_type OK TOKEN_CLS
95
+ lora.modules_to_save OK None
96
+ lora.init_lora_weights DEFAULT True
97
+ lora.use_rslora DEFAULT False
98
+ lora.fan_in_fan_out DEFAULT False
99
+ environment.push_to_hub OK True
100
+ environment.hub_model_id OK arthrod/gliner-pii-token-multipii-v1-spam-v1
101
+ environment.hf_token OK None
102
+ environment.report_to OK wandb
103
+ environment.wandb_project OK gliner-pii
104
+ environment.wandb_entity OK None
105
+ environment.wandb_api_key OK None
106
+ environment.cuda_visible_devices OK 0
107
+
108
+ WARNINGS:
109
+ [DEFAULT] 'model.labels_encoder_config' not set -- using default: None
110
+ [DEFAULT] 'model.labels_decoder_config' not set -- using default: None
111
+ [DEFAULT] 'training.gradient_checkpointing' not set -- using default: False
112
+ [DEFAULT] 'training.eval_steps' not set -- using default: None
113
+ [DEFAULT] 'training.eval_on_start' not set -- using default: False
114
+ [DEFAULT] 'lora.init_lora_weights' not set -- using default: True
115
+ [DEFAULT] 'lora.use_rslora' not set -- using default: False
116
+ [DEFAULT] 'lora.fan_in_fan_out' not set -- using default: False
117
+
118
+ Total: 103 fields | 8 warnings | 0 errors
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:506beabd886b40a1fe008af4ea41be28e5a723410e5bc6ccc18850b1fea61bad
3
+ size 16034669
tokenizer_config.json ADDED
@@ -0,0 +1,123 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": true,
3
+ "backend": "tokenizers",
4
+ "bos_token": "[CLS]",
5
+ "cls_token": "[CLS]",
6
+ "do_lower_case": false,
7
+ "eos_token": "[SEP]",
8
+ "extra_special_tokens": [
9
+ "[PAD]",
10
+ "[CLS]",
11
+ "[SEP]",
12
+ "▁<extra_id_99>",
13
+ "▁<extra_id_98>",
14
+ "▁<extra_id_97>",
15
+ "▁<extra_id_96>",
16
+ "▁<extra_id_95>",
17
+ "▁<extra_id_94>",
18
+ "▁<extra_id_93>",
19
+ "▁<extra_id_92>",
20
+ "▁<extra_id_91>",
21
+ "▁<extra_id_90>",
22
+ "▁<extra_id_89>",
23
+ "▁<extra_id_88>",
24
+ "▁<extra_id_87>",
25
+ "▁<extra_id_86>",
26
+ "▁<extra_id_85>",
27
+ "▁<extra_id_84>",
28
+ "▁<extra_id_83>",
29
+ "▁<extra_id_82>",
30
+ "▁<extra_id_81>",
31
+ "▁<extra_id_80>",
32
+ "▁<extra_id_79>",
33
+ "▁<extra_id_78>",
34
+ "▁<extra_id_77>",
35
+ "▁<extra_id_76>",
36
+ "▁<extra_id_75>",
37
+ "▁<extra_id_74>",
38
+ "▁<extra_id_73>",
39
+ "▁<extra_id_72>",
40
+ "▁<extra_id_71>",
41
+ "▁<extra_id_70>",
42
+ "▁<extra_id_69>",
43
+ "▁<extra_id_68>",
44
+ "▁<extra_id_67>",
45
+ "▁<extra_id_66>",
46
+ "▁<extra_id_65>",
47
+ "▁<extra_id_64>",
48
+ "▁<extra_id_63>",
49
+ "▁<extra_id_62>",
50
+ "▁<extra_id_61>",
51
+ "▁<extra_id_60>",
52
+ "▁<extra_id_59>",
53
+ "▁<extra_id_58>",
54
+ "▁<extra_id_57>",
55
+ "▁<extra_id_56>",
56
+ "▁<extra_id_55>",
57
+ "▁<extra_id_54>",
58
+ "▁<extra_id_53>",
59
+ "▁<extra_id_52>",
60
+ "▁<extra_id_51>",
61
+ "▁<extra_id_50>",
62
+ "▁<extra_id_49>",
63
+ "▁<extra_id_48>",
64
+ "▁<extra_id_47>",
65
+ "▁<extra_id_46>",
66
+ "▁<extra_id_45>",
67
+ "▁<extra_id_44>",
68
+ "▁<extra_id_43>",
69
+ "▁<extra_id_42>",
70
+ "▁<extra_id_41>",
71
+ "▁<extra_id_40>",
72
+ "▁<extra_id_39>",
73
+ "▁<extra_id_38>",
74
+ "▁<extra_id_37>",
75
+ "▁<extra_id_36>",
76
+ "▁<extra_id_35>",
77
+ "▁<extra_id_34>",
78
+ "▁<extra_id_33>",
79
+ "▁<extra_id_32>",
80
+ "▁<extra_id_31>",
81
+ "▁<extra_id_30>",
82
+ "▁<extra_id_29>",
83
+ "▁<extra_id_28>",
84
+ "▁<extra_id_27>",
85
+ "▁<extra_id_26>",
86
+ "▁<extra_id_25>",
87
+ "▁<extra_id_24>",
88
+ "▁<extra_id_23>",
89
+ "▁<extra_id_22>",
90
+ "▁<extra_id_21>",
91
+ "▁<extra_id_20>",
92
+ "▁<extra_id_19>",
93
+ "▁<extra_id_18>",
94
+ "▁<extra_id_17>",
95
+ "▁<extra_id_16>",
96
+ "▁<extra_id_15>",
97
+ "▁<extra_id_14>",
98
+ "▁<extra_id_13>",
99
+ "▁<extra_id_12>",
100
+ "▁<extra_id_11>",
101
+ "▁<extra_id_10>",
102
+ "▁<extra_id_9>",
103
+ "▁<extra_id_8>",
104
+ "▁<extra_id_7>",
105
+ "▁<extra_id_6>",
106
+ "▁<extra_id_5>",
107
+ "▁<extra_id_4>",
108
+ "▁<extra_id_3>",
109
+ "▁<extra_id_2>",
110
+ "▁<extra_id_1>",
111
+ "▁<extra_id_0>"
112
+ ],
113
+ "is_local": false,
114
+ "mask_token": "[MASK]",
115
+ "model_max_length": 1000000000000000019884624838656,
116
+ "pad_token": "[PAD]",
117
+ "sep_token": "[SEP]",
118
+ "split_by_punct": false,
119
+ "tokenizer_class": "DebertaV2Tokenizer",
120
+ "unk_id": 3,
121
+ "unk_token": "[UNK]",
122
+ "vocab_type": "spm"
123
+ }
training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:56e7a494ca68f698936f2394ec3e507d8cb5d23b525b61a2ee8d69d800f92cc7
3
+ size 5585
validation_20260224T084509Z.log ADDED
@@ -0,0 +1,260 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-02-24 08:45:09,750 | INFO | Log file: runs/b-token-v1/validation_20260224T084509Z.log
2
+ 2026-02-24 08:45:09,751 | INFO | Timestamp: 20260224T084509Z
3
+ 2026-02-24 08:45:09,751 | INFO | [OK] 'run.name' = 'b-gliner-pii-token-multipii-v1-spam-v1'
4
+ 2026-02-24 08:45:09,752 | INFO | [OK] 'run.description' = 'Domain adaptation for PII-in-spam using token-level GLiNER starting from gliner_multi_pii-v1.'
5
+ 2026-02-24 08:45:09,752 | INFO | [OK] 'run.tags' = ['pii', 'spam', 'gliner', 'token_level', 'domain-adapt', 'multipii-v1']
6
+ 2026-02-24 08:45:09,753 | INFO | [OK] 'run.seed' = 42
7
+ 2026-02-24 08:45:09,754 | INFO | [OK] 'model.model_name' = 'microsoft/deberta-v3-base'
8
+ 2026-02-24 08:45:09,754 | INFO | [OK] 'model.name' = 'gliner-pii-token-multipii-v1-spam'
9
+ 2026-02-24 08:45:09,755 | INFO | [OK] 'model.labels_encoder' = None
10
+ 2026-02-24 08:45:09,755 | INFO | [OK] 'model.labels_decoder' = None
11
+ 2026-02-24 08:45:09,755 | INFO | [OK] 'model.decoder_mode' = None
12
+ 2026-02-24 08:45:09,756 | INFO | [OK] 'model.full_decoder_context' = True
13
+ 2026-02-24 08:45:09,756 | INFO | [OK] 'model.blank_entity_prob' = 0.1
14
+ 2026-02-24 08:45:09,757 | INFO | [OK] 'model.decoder_loss_coef' = 0.5
15
+ 2026-02-24 08:45:09,757 | INFO | [OK] 'model.relations_layer' = None
16
+ 2026-02-24 08:45:09,758 | INFO | [OK] 'model.triples_layer' = None
17
+ 2026-02-24 08:45:09,758 | INFO | [OK] 'model.embed_rel_token' = True
18
+ 2026-02-24 08:45:09,758 | INFO | [OK] 'model.rel_token_index' = -1
19
+ 2026-02-24 08:45:09,759 | INFO | [OK] 'model.rel_token' = '<<REL>>'
20
+ 2026-02-24 08:45:09,759 | INFO | [OK] 'model.adjacency_loss_coef' = 1.0
21
+ 2026-02-24 08:45:09,760 | INFO | [OK] 'model.relation_loss_coef' = 1.0
22
+ 2026-02-24 08:45:09,760 | INFO | [OK] 'model.span_mode' = 'token_level'
23
+ 2026-02-24 08:45:09,761 | INFO | [OK] 'model.max_width' = 12
24
+ 2026-02-24 08:45:09,761 | INFO | [OK] 'model.represent_spans' = False
25
+ 2026-02-24 08:45:09,761 | INFO | [OK] 'model.neg_spans_ratio' = 1.0
26
+ 2026-02-24 08:45:09,762 | INFO | [OK] 'model.hidden_size' = 768
27
+ 2026-02-24 08:45:09,762 | INFO | [OK] 'model.dropout' = 0.25
28
+ 2026-02-24 08:45:09,763 | INFO | [OK] 'model.fine_tune' = True
29
+ 2026-02-24 08:45:09,763 | INFO | [OK] 'model.subtoken_pooling' = 'first'
30
+ 2026-02-24 08:45:09,763 | INFO | [OK] 'model.fuse_layers' = False
31
+ 2026-02-24 08:45:09,764 | INFO | [OK] 'model.post_fusion_schema' = ''
32
+ 2026-02-24 08:45:09,764 | INFO | [OK] 'model.num_post_fusion_layers' = 1
33
+ 2026-02-24 08:45:09,765 | INFO | [OK] 'model.num_rnn_layers' = 0
34
+ 2026-02-24 08:45:09,765 | INFO | [OK] 'model.max_len' = 384
35
+ 2026-02-24 08:45:09,765 | INFO | [OK] 'model.max_types' = 100
36
+ 2026-02-24 08:45:09,766 | INFO | [OK] 'model.max_neg_type_ratio' = 2
37
+ 2026-02-24 08:45:09,766 | INFO | [OK] 'model.words_splitter_type' = 'whitespace'
38
+ 2026-02-24 08:45:09,767 | INFO | [OK] 'model.embed_ent_token' = True
39
+ 2026-02-24 08:45:09,767 | INFO | [OK] 'model.class_token_index' = -1
40
+ 2026-02-24 08:45:09,768 | INFO | [OK] 'model.ent_token' = '<<ENT>>'
41
+ 2026-02-24 08:45:09,768 | INFO | [OK] 'model.sep_token' = '<<SEP>>'
42
+ 2026-02-24 08:45:09,768 | INFO | [OK] 'model.token_loss_coef' = 1.0
43
+ 2026-02-24 08:45:09,769 | INFO | [OK] 'model.span_loss_coef' = 1.0
44
+ 2026-02-24 08:45:09,769 | INFO | [OK] 'model.encoder_config' = None
45
+ 2026-02-24 08:45:09,770 | WARNING | [DEFAULT] 'model.labels_encoder_config' not set -- using default: None
46
+ 2026-02-24 08:45:09,770 | WARNING | [DEFAULT] 'model.labels_decoder_config' not set -- using default: None
47
+ 2026-02-24 08:45:09,771 | INFO | [OK] 'model._attn_implementation' = 'flash_attention_2'
48
+ 2026-02-24 08:45:09,771 | INFO | [OK] 'model.vocab_size' = -1
49
+ 2026-02-24 08:45:09,772 | INFO | [OK] 'data.root_dir' = 'gliner_logs/gliner-pii-token-multipii-v1-spam-v1'
50
+ 2026-02-24 08:45:09,772 | INFO | [OK] 'data.train_data' = 'arthrod/gliner-flex-pii-ready-v5::train'
51
+ 2026-02-24 08:45:09,772 | INFO | [OK] 'data.val_data_dir' = 'arthrod/gliner-flex-pii-ready-v5::eval'
52
+ 2026-02-24 08:45:09,773 | INFO | [OK] 'training.prev_path' = 'urchade/gliner_multi_pii-v1'
53
+ 2026-02-24 08:45:09,773 | INFO | [OK] 'training.num_steps' = 12000
54
+ 2026-02-24 08:45:09,774 | INFO | [OK] 'training.scheduler_type' = 'cosine'
55
+ 2026-02-24 08:45:09,774 | INFO | [OK] 'training.warmup_ratio' = 0.05
56
+ 2026-02-24 08:45:09,775 | INFO | [OK] 'training.train_batch_size' = 32
57
+ 2026-02-24 08:45:09,775 | INFO | [OK] 'training.eval_batch_size' = 64
58
+ 2026-02-24 08:45:09,775 | INFO | [OK] 'training.gradient_accumulation_steps' = 1
59
+ 2026-02-24 08:45:09,776 | INFO | [OK] 'training.max_grad_norm' = 1.0
60
+ 2026-02-24 08:45:09,776 | INFO | [OK] 'training.optimizer' = 'adamw_torch'
61
+ 2026-02-24 08:45:09,777 | INFO | [OK] 'training.lr_encoder' = 8e-06
62
+ 2026-02-24 08:45:09,777 | INFO | [OK] 'training.lr_others' = 4e-05
63
+ 2026-02-24 08:45:09,778 | INFO | [OK] 'training.weight_decay_encoder' = 0.01
64
+ 2026-02-24 08:45:09,778 | INFO | [OK] 'training.weight_decay_other' = 0.01
65
+ 2026-02-24 08:45:09,778 | INFO | [OK] 'training.loss_alpha' = 0.75
66
+ 2026-02-24 08:45:09,779 | INFO | [OK] 'training.loss_gamma' = 2.0
67
+ 2026-02-24 08:45:09,779 | INFO | [OK] 'training.loss_prob_margin' = 0.0
68
+ 2026-02-24 08:45:09,780 | INFO | [OK] 'training.label_smoothing' = 0.0
69
+ 2026-02-24 08:45:09,780 | INFO | [OK] 'training.loss_reduction' = 'sum'
70
+ 2026-02-24 08:45:09,781 | INFO | [OK] 'training.negatives' = 2.0
71
+ 2026-02-24 08:45:09,781 | INFO | [OK] 'training.masking' = 'none'
72
+ 2026-02-24 08:45:09,781 | INFO | [OK] 'training.eval_every' = 500
73
+ 2026-02-24 08:45:09,782 | INFO | [OK] 'training.save_total_limit' = 20
74
+ 2026-02-24 08:45:09,782 | INFO | [OK] 'training.logging_steps' = 10
75
+ 2026-02-24 08:45:09,783 | INFO | [OK] 'training.bf16' = True
76
+ 2026-02-24 08:45:09,783 | INFO | [OK] 'training.fp16' = False
77
+ 2026-02-24 08:45:09,783 | INFO | [OK] 'training.use_cpu' = False
78
+ 2026-02-24 08:45:09,784 | INFO | [OK] 'training.dataloader_num_workers' = 18
79
+ 2026-02-24 08:45:09,784 | INFO | [OK] 'training.dataloader_pin_memory' = True
80
+ 2026-02-24 08:45:09,785 | INFO | [OK] 'training.dataloader_persistent_workers' = False
81
+ 2026-02-24 08:45:09,785 | INFO | [OK] 'training.dataloader_prefetch_factor' = 2
82
+ 2026-02-24 08:45:09,786 | INFO | [OK] 'training.freeze_components' = None
83
+ 2026-02-24 08:45:09,786 | WARNING | [DEFAULT] 'training.gradient_checkpointing' not set -- using default: False
84
+ 2026-02-24 08:45:09,786 | WARNING | [DEFAULT] 'training.eval_steps' not set -- using default: None
85
+ 2026-02-24 08:45:09,787 | WARNING | [DEFAULT] 'training.eval_on_start' not set -- using default: False
86
+ 2026-02-24 08:45:09,787 | INFO | [OK] 'training.compile_model' = False
87
+ 2026-02-24 08:45:09,788 | INFO | [OK] 'lora.enabled' = False
88
+ 2026-02-24 08:45:09,788 | INFO | [OK] 'lora.r' = 16
89
+ 2026-02-24 08:45:09,789 | INFO | [OK] 'lora.lora_alpha' = 32
90
+ 2026-02-24 08:45:09,789 | INFO | [OK] 'lora.lora_dropout' = 0.05
91
+ 2026-02-24 08:45:09,789 | INFO | [OK] 'lora.bias' = 'none'
92
+ 2026-02-24 08:45:09,790 | INFO | [OK] 'lora.target_modules' = ['all-linear']
93
+ 2026-02-24 08:45:09,790 | INFO | [OK] 'lora.task_type' = 'TOKEN_CLS'
94
+ 2026-02-24 08:45:09,791 | INFO | [OK] 'lora.modules_to_save' = None
95
+ 2026-02-24 08:45:09,791 | WARNING | [DEFAULT] 'lora.init_lora_weights' not set -- using default: True
96
+ 2026-02-24 08:45:09,792 | WARNING | [DEFAULT] 'lora.use_rslora' not set -- using default: False
97
+ 2026-02-24 08:45:09,792 | WARNING | [DEFAULT] 'lora.fan_in_fan_out' not set -- using default: False
98
+ 2026-02-24 08:45:09,793 | INFO | [OK] 'environment.push_to_hub' = True
99
+ 2026-02-24 08:45:09,793 | INFO | [OK] 'environment.hub_model_id' = 'arthrod/gliner-pii-token-multipii-v1-spam-v1'
100
+ 2026-02-24 08:45:09,793 | INFO | [OK] 'environment.hf_token' = None
101
+ 2026-02-24 08:45:09,794 | INFO | [OK] 'environment.report_to' = 'wandb'
102
+ 2026-02-24 08:45:09,794 | INFO | [OK] 'environment.wandb_project' = 'gliner-pii'
103
+ 2026-02-24 08:45:09,795 | INFO | [OK] 'environment.wandb_entity' = None
104
+ 2026-02-24 08:45:09,795 | INFO | [OK] 'environment.wandb_api_key' = None
105
+ 2026-02-24 08:45:09,796 | INFO | [OK] 'environment.cuda_visible_devices' = '0'
106
+ 2026-02-24 08:45:09,796 | INFO | [DATA] Training source is HF dataset repo 'arthrod/gliner-flex-pii-ready-v5' (split='train')
107
+ 2026-02-24 08:45:09,797 | INFO | [DATA] Validation source is HF dataset repo 'arthrod/gliner-flex-pii-ready-v5' (split='eval')
108
+ 2026-02-24 08:45:09,960 | INFO | [HF] Authenticated as 'arthrod'
109
+ 2026-02-24 08:45:10,019 | INFO | [WANDB] Authenticated as 'arthrod'
110
+ 2026-02-24 08:45:10,039 | INFO | Summary written to log file
111
+ 2026-02-24 08:45:10,040 | INFO | Summary saved to runs/b-token-v1/summary_20260224T084509Z.txt
112
+ 2026-02-24 08:45:10,043 | INFO | Resolved config saved to runs/b-token-v1/config.yaml
113
+ 2026-02-24 08:45:10,043 | INFO | Preparing training run ...
114
+ 2026-02-24 08:45:13,463 | INFO | Loading pretrained model from: urchade/gliner_multi_pii-v1
115
+ 2026-02-24 08:45:26,843 | INFO | Model class: UniEncoderSpanGLiNER
116
+ 2026-02-24 08:45:26,844 | INFO | [CONFIG] Overrode pretrained config with YAML values:
117
+ 2026-02-24 08:45:26,844 | INFO | max_types: 25 → 100
118
+ 2026-02-24 08:45:26,845 | INFO | max_neg_type_ratio: 3 → 2
119
+ 2026-02-24 08:45:26,846 | INFO | dropout: 0.4 → 0.25
120
+ 2026-02-24 08:45:26,846 | INFO | Loading training data from HF dataset 'arthrod/gliner-flex-pii-ready-v5' (split='train')
121
+ 2026-02-24 08:47:25,793 | INFO | Training samples: 1111948
122
+ 2026-02-24 08:47:25,794 | INFO | Loading validation data from HF dataset 'arthrod/gliner-flex-pii-ready-v5' (split='eval')
123
+ 2026-02-24 08:48:30,030 | INFO | Validation samples: 526449
124
+ 2026-02-24 08:48:30,032 | INFO | ============================================================
125
+ 2026-02-24 08:48:30,032 | INFO | [DIAG] Model architecture summary
126
+ 2026-02-24 08:48:30,033 | INFO | [DIAG] Class: UniEncoderSpanGLiNER
127
+ 2026-02-24 08:48:30,033 | INFO | [DIAG] Total params: 288,949,504
128
+ 2026-02-24 08:48:30,034 | INFO | [DIAG] Trainable params: 288,949,504 (100.00%)
129
+ 2026-02-24 08:48:30,034 | INFO | [DIAG] Dataset sizes
130
+ 2026-02-24 08:48:30,035 | INFO | [DIAG] Train: 1111948
131
+ 2026-02-24 08:48:30,035 | INFO | [DIAG] Eval: 526449
132
+ 2026-02-24 08:48:30,036 | INFO | [DIAG] Model config snapshot
133
+ 2026-02-24 08:48:30,036 | INFO | [DIAG] return_dict: True
134
+ 2026-02-24 08:48:30,037 | INFO | [DIAG] output_hidden_states: False
135
+ 2026-02-24 08:48:30,038 | INFO | [DIAG] dtype: None
136
+ 2026-02-24 08:48:30,038 | INFO | [DIAG] chunk_size_feed_forward: 0
137
+ 2026-02-24 08:48:30,039 | INFO | [DIAG] is_encoder_decoder: False
138
+ 2026-02-24 08:48:30,039 | INFO | [DIAG] architectures: None
139
+ 2026-02-24 08:48:30,040 | INFO | [DIAG] id2label: {0: 'LABEL_0', 1: 'LABEL_1'}
140
+ 2026-02-24 08:48:30,040 | INFO | [DIAG] label2id: {'LABEL_0': 0, 'LABEL_1': 1}
141
+ 2026-02-24 08:48:30,041 | INFO | [DIAG] problem_type: None
142
+ 2026-02-24 08:48:30,041 | INFO | [DIAG] transformers_version: None
143
+ 2026-02-24 08:48:30,041 | INFO | [DIAG] size_sup: -1
144
+ 2026-02-24 08:48:30,042 | INFO | [DIAG] shuffle_types: True
145
+ 2026-02-24 08:48:30,042 | INFO | [DIAG] random_drop: False
146
+ 2026-02-24 08:48:30,043 | INFO | [DIAG] lr_encoder: '5e-6'
147
+ 2026-02-24 08:48:30,043 | INFO | [DIAG] lr_others: '1e-5'
148
+ 2026-02-24 08:48:30,044 | INFO | [DIAG] num_steps: 30000
149
+ 2026-02-24 08:48:30,044 | INFO | [DIAG] warmup_ratio: 5000
150
+ 2026-02-24 08:48:30,045 | INFO | [DIAG] train_batch_size: 8
151
+ 2026-02-24 08:48:30,045 | INFO | [DIAG] eval_every: 5000
152
+ 2026-02-24 08:48:30,046 | INFO | [DIAG] encoder_config: DebertaV2Config {
153
+ "attention_probs_dropout_prob": 0.1,
154
+ "bos_token_id": null,
155
+ "eos_token_id": null,
156
+ "hidden_act": "gelu",
157
+ "hidden_dropout_prob": 0.1,
158
+ "hidden_size": 768,
159
+ "initializer_range": 0.02,
160
+ "intermediate_size": 3072,
161
+ "layer_norm_eps": 1e-07,
162
+ "legacy": true,
163
+ "max_position_embeddings": 512,
164
+ "max_relative_positions": -1,
165
+ "model_type": "deberta-v2",
166
+ "norm_rel_ebd": "layer_norm",
167
+ "num_attention_heads": 12,
168
+ "num_hidden_layers": 12,
169
+ "pad_token_id": 0,
170
+ "pooler_dropout": 0,
171
+ "pooler_hidden_act": "gelu",
172
+ "pooler_hidden_size": 768,
173
+ "pos_att_type": [
174
+ "p2c",
175
+ "c2p"
176
+ ],
177
+ "position_biased_input": false,
178
+ "position_buckets": 256,
179
+ "relative_attention": true,
180
+ "share_att_key": true,
181
+ "tie_word_embeddings": true,
182
+ "transformers_version": "5.1.0",
183
+ "type_vocab_size": 0,
184
+ "vocab_size": 250105
185
+ }
186
+
187
+ 2026-02-24 08:48:30,049 | INFO | [DIAG] model_name: 'microsoft/mdeberta-v3-base'
188
+ 2026-02-24 08:48:30,049 | INFO | [DIAG] name: 'correct'
189
+ 2026-02-24 08:48:30,050 | INFO | [DIAG] max_width: 12
190
+ 2026-02-24 08:48:30,050 | INFO | [DIAG] hidden_size: 512
191
+ 2026-02-24 08:48:30,051 | INFO | [DIAG] dropout: 0.25
192
+ 2026-02-24 08:48:30,051 | INFO | [DIAG] fine_tune: True
193
+ 2026-02-24 08:48:30,052 | INFO | [DIAG] subtoken_pooling: 'first'
194
+ 2026-02-24 08:48:30,052 | INFO | [DIAG] span_mode: 'markerV0'
195
+ 2026-02-24 08:48:30,053 | INFO | [DIAG] post_fusion_schema: ''
196
+ 2026-02-24 08:48:30,053 | INFO | [DIAG] num_post_fusion_layers: 1
197
+ 2026-02-24 08:48:30,054 | INFO | [DIAG] vocab_size: 250105
198
+ 2026-02-24 08:48:30,054 | INFO | [DIAG] max_neg_type_ratio: 2
199
+ 2026-02-24 08:48:30,055 | INFO | [DIAG] max_types: 100
200
+ 2026-02-24 08:48:30,055 | INFO | [DIAG] max_len: 384
201
+ 2026-02-24 08:48:30,055 | INFO | [DIAG] words_splitter_type: 'whitespace'
202
+ 2026-02-24 08:48:30,056 | INFO | [DIAG] num_rnn_layers: 1
203
+ 2026-02-24 08:48:30,056 | INFO | [DIAG] fuse_layers: False
204
+ 2026-02-24 08:48:30,057 | INFO | [DIAG] class_token_index: 250103
205
+ 2026-02-24 08:48:30,057 | INFO | [DIAG] embed_ent_token: True
206
+ 2026-02-24 08:48:30,058 | INFO | [DIAG] ent_token: '<<ENT>>'
207
+ 2026-02-24 08:48:30,058 | INFO | [DIAG] sep_token: '<<SEP>>'
208
+ 2026-02-24 08:48:30,059 | INFO | [DIAG] token_loss_coef: 1.0
209
+ 2026-02-24 08:48:30,059 | INFO | [DIAG] span_loss_coef: 1.0
210
+ 2026-02-24 08:48:30,060 | INFO | [DIAG] represent_spans: False
211
+ 2026-02-24 08:48:30,060 | INFO | [DIAG] neg_spans_ratio: 1.0
212
+ 2026-02-24 08:48:30,061 | INFO | [DIAG] model_type: 'gliner_uni_encoder_span'
213
+ 2026-02-24 08:48:30,061 | INFO | [DIAG] trust_remote_code: False
214
+ 2026-02-24 08:48:30,061 | INFO | [DIAG] Training kwargs (non-dataset)
215
+ 2026-02-24 08:48:30,062 | INFO | [DIAG] output_dir: 'runs/b-token-v1'
216
+ 2026-02-24 08:48:30,062 | INFO | [DIAG] freeze_components: None
217
+ 2026-02-24 08:48:30,063 | INFO | [DIAG] compile_model: False
218
+ 2026-02-24 08:48:30,063 | INFO | [DIAG] resume_from_checkpoint: None
219
+ 2026-02-24 08:48:30,064 | INFO | [DIAG] max_steps: 12000
220
+ 2026-02-24 08:48:30,064 | INFO | [DIAG] lr_scheduler_type: 'cosine'
221
+ 2026-02-24 08:48:30,064 | INFO | [DIAG] warmup_ratio: 0.05
222
+ 2026-02-24 08:48:30,065 | INFO | [DIAG] per_device_train_batch_size: 32
223
+ 2026-02-24 08:48:30,065 | INFO | [DIAG] per_device_eval_batch_size: 64
224
+ 2026-02-24 08:48:30,066 | INFO | [DIAG] learning_rate: 8e-06
225
+ 2026-02-24 08:48:30,066 | INFO | [DIAG] others_lr: 4e-05
226
+ 2026-02-24 08:48:30,067 | INFO | [DIAG] weight_decay: 0.01
227
+ 2026-02-24 08:48:30,067 | INFO | [DIAG] others_weight_decay: 0.01
228
+ 2026-02-24 08:48:30,068 | INFO | [DIAG] max_grad_norm: 1.0
229
+ 2026-02-24 08:48:30,068 | INFO | [DIAG] optim: 'adamw_torch'
230
+ 2026-02-24 08:48:30,068 | INFO | [DIAG] focal_loss_alpha: 0.75
231
+ 2026-02-24 08:48:30,069 | INFO | [DIAG] focal_loss_gamma: 2.0
232
+ 2026-02-24 08:48:30,069 | INFO | [DIAG] focal_loss_prob_margin: 0.0
233
+ 2026-02-24 08:48:30,070 | INFO | [DIAG] label_smoothing: 0.0
234
+ 2026-02-24 08:48:30,070 | INFO | [DIAG] loss_reduction: 'sum'
235
+ 2026-02-24 08:48:30,071 | INFO | [DIAG] negatives: 2.0
236
+ 2026-02-24 08:48:30,071 | INFO | [DIAG] masking: 'none'
237
+ 2026-02-24 08:48:30,071 | INFO | [DIAG] save_steps: 500
238
+ 2026-02-24 08:48:30,072 | INFO | [DIAG] logging_steps: 10
239
+ 2026-02-24 08:48:30,072 | INFO | [DIAG] save_total_limit: 20
240
+ 2026-02-24 08:48:30,073 | INFO | [DIAG] eval_strategy: 'steps'
241
+ 2026-02-24 08:48:30,073 | INFO | [DIAG] eval_steps: 500
242
+ 2026-02-24 08:48:30,074 | INFO | [DIAG] eval_on_start: False
243
+ 2026-02-24 08:48:30,074 | INFO | [DIAG] bf16: True
244
+ 2026-02-24 08:48:30,074 | INFO | [DIAG] fp16: False
245
+ 2026-02-24 08:48:30,075 | INFO | [DIAG] gradient_checkpointing: False
246
+ 2026-02-24 08:48:30,075 | INFO | [DIAG] use_cpu: False
247
+ 2026-02-24 08:48:30,076 | INFO | [DIAG] dataloader_num_workers: 18
248
+ 2026-02-24 08:48:30,076 | INFO | [DIAG] dataloader_pin_memory: True
249
+ 2026-02-24 08:48:30,077 | INFO | [DIAG] dataloader_persistent_workers: False
250
+ 2026-02-24 08:48:30,077 | INFO | [DIAG] dataloader_prefetch_factor: 2
251
+ 2026-02-24 08:48:30,077 | INFO | [DIAG] report_to: 'wandb'
252
+ 2026-02-24 08:48:30,078 | INFO | [DIAG] run_name: 'b-gliner-pii-token-multipii-v1-spam-v1'
253
+ 2026-02-24 08:48:30,078 | INFO | [DIAG] gradient_accumulation_steps: 1
254
+ 2026-02-24 08:48:30,079 | INFO | [DIAG] seed: 42
255
+ 2026-02-24 08:48:30,079 | INFO | [DIAG] remove_unused_columns: False
256
+ 2026-02-24 08:48:30,080 | INFO | [DIAG] push_to_hub: True
257
+ 2026-02-24 08:48:30,080 | INFO | [DIAG] hub_model_id: 'arthrod/gliner-pii-token-multipii-v1-spam-v1'
258
+ 2026-02-24 08:48:30,081 | INFO | [DIAG] lora_config: None
259
+ 2026-02-24 08:48:30,081 | INFO | ============================================================
260
+ 2026-02-24 08:48:30,082 | INFO | Starting training ...