Rogendo commited on
Commit
1690859
·
verified ·
1 Parent(s): ee6f483

Upload JengaAI trained multi-task NLP model

Browse files
checkpoints/best/checkpoint_meta.json ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 3,
3
+ "metrics": {
4
+ "fraud_detection_fraud_accuracy": 1.0,
5
+ "fraud_detection_fraud_precision": 1.0,
6
+ "fraud_detection_fraud_recall": 1.0,
7
+ "fraud_detection_fraud_f1": 1.0,
8
+ "sentiment_analysis_sentiment_accuracy": 0.3333333333333333,
9
+ "sentiment_analysis_sentiment_precision": 0.1111111111111111,
10
+ "sentiment_analysis_sentiment_recall": 0.3333333333333333,
11
+ "sentiment_analysis_sentiment_f1": 0.16666666666666666,
12
+ "call_quality_opening_precision": 0.967,
13
+ "call_quality_opening_recall": 0.967,
14
+ "call_quality_opening_f1": 0.967,
15
+ "call_quality_listening_precision": 0.8930771125893077,
16
+ "call_quality_listening_recall": 0.9526938239159002,
17
+ "call_quality_listening_f1": 0.9219226856561547,
18
+ "call_quality_proactiveness_precision": 0.7458045409674235,
19
+ "call_quality_proactiveness_recall": 0.8683908045977011,
20
+ "call_quality_proactiveness_f1": 0.8024429102496017,
21
+ "call_quality_resolution_precision": 0.9176672384219554,
22
+ "call_quality_resolution_recall": 0.8984047019311503,
23
+ "call_quality_resolution_f1": 0.9079338141705557,
24
+ "call_quality_hold_precision": 0.856,
25
+ "call_quality_hold_recall": 0.5194174757281553,
26
+ "call_quality_hold_f1": 0.6465256797583081,
27
+ "call_quality_closing_precision": 0.881,
28
+ "call_quality_closing_recall": 0.881,
29
+ "call_quality_closing_f1": 0.881,
30
+ "eval_loss": 1.947792258629432,
31
+ "train_loss_avg": 1.878378942786228
32
+ },
33
+ "is_best": true
34
+ }
checkpoints/best/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:61a5c55199089517c91bbbcc1e47ee321da91656ea2162a45c4a89a03bcf9dab
3
+ size 270300931
checkpoints/epoch_7/checkpoint_meta.json ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 7,
3
+ "metrics": {
4
+ "fraud_detection_fraud_accuracy": 1.0,
5
+ "fraud_detection_fraud_precision": 1.0,
6
+ "fraud_detection_fraud_recall": 1.0,
7
+ "fraud_detection_fraud_f1": 1.0,
8
+ "sentiment_analysis_sentiment_accuracy": 0.3333333333333333,
9
+ "sentiment_analysis_sentiment_precision": 0.16666666666666666,
10
+ "sentiment_analysis_sentiment_recall": 0.3333333333333333,
11
+ "sentiment_analysis_sentiment_f1": 0.2222222222222222,
12
+ "call_quality_opening_precision": 0.965,
13
+ "call_quality_opening_recall": 0.965,
14
+ "call_quality_opening_f1": 0.965,
15
+ "call_quality_listening_precision": 0.9060505488894562,
16
+ "call_quality_listening_recall": 0.9327201051248357,
17
+ "call_quality_listening_f1": 0.9191919191919192,
18
+ "call_quality_proactiveness_precision": 0.7565891472868217,
19
+ "call_quality_proactiveness_recall": 0.8413793103448276,
20
+ "call_quality_proactiveness_f1": 0.796734693877551,
21
+ "call_quality_resolution_precision": 0.9000273897562312,
22
+ "call_quality_resolution_recall": 0.9196753428491464,
23
+ "call_quality_resolution_f1": 0.9097452934662237,
24
+ "call_quality_hold_precision": 0.7371428571428571,
25
+ "call_quality_hold_recall": 0.6262135922330098,
26
+ "call_quality_hold_f1": 0.6771653543307087,
27
+ "call_quality_closing_precision": 0.869,
28
+ "call_quality_closing_recall": 0.869,
29
+ "call_quality_closing_f1": 0.869,
30
+ "eval_loss": 2.056771606665391,
31
+ "train_loss_avg": 1.4706985300465945
32
+ },
33
+ "is_best": false
34
+ }
checkpoints/epoch_7/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:61273ac9e032dea822ee629ab27acb039dabe6bb4542926f19455bed09369b2d
3
+ size 270300931
checkpoints/epoch_8/checkpoint_meta.json ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 8,
3
+ "metrics": {
4
+ "fraud_detection_fraud_accuracy": 1.0,
5
+ "fraud_detection_fraud_precision": 1.0,
6
+ "fraud_detection_fraud_recall": 1.0,
7
+ "fraud_detection_fraud_f1": 1.0,
8
+ "sentiment_analysis_sentiment_accuracy": 0.3333333333333333,
9
+ "sentiment_analysis_sentiment_precision": 0.16666666666666666,
10
+ "sentiment_analysis_sentiment_recall": 0.3333333333333333,
11
+ "sentiment_analysis_sentiment_f1": 0.2222222222222222,
12
+ "call_quality_opening_precision": 0.964,
13
+ "call_quality_opening_recall": 0.964,
14
+ "call_quality_opening_f1": 0.964,
15
+ "call_quality_listening_precision": 0.9018963337547409,
16
+ "call_quality_listening_recall": 0.9374507227332457,
17
+ "call_quality_listening_f1": 0.9193298969072164,
18
+ "call_quality_proactiveness_precision": 0.7551020408163265,
19
+ "call_quality_proactiveness_recall": 0.8293103448275863,
20
+ "call_quality_proactiveness_f1": 0.790468364831553,
21
+ "call_quality_resolution_precision": 0.9021229666390956,
22
+ "call_quality_resolution_recall": 0.9157570668905681,
23
+ "call_quality_resolution_f1": 0.9088888888888889,
24
+ "call_quality_hold_precision": 0.7575757575757576,
25
+ "call_quality_hold_recall": 0.6067961165048543,
26
+ "call_quality_hold_f1": 0.6738544474393531,
27
+ "call_quality_closing_precision": 0.869,
28
+ "call_quality_closing_recall": 0.869,
29
+ "call_quality_closing_f1": 0.869,
30
+ "eval_loss": 2.068362368643284,
31
+ "train_loss_avg": 1.4034666525022614
32
+ },
33
+ "is_best": false
34
+ }
checkpoints/epoch_8/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3a06d1dc18f367e75599ea712ea481525e701f446c2660ba016f124e6f67746a
3
+ size 270300931
encoder_config/config.json ADDED
@@ -0,0 +1,25 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "activation": "gelu",
3
+ "architectures": [
4
+ "DistilBertForMaskedLM"
5
+ ],
6
+ "attention_dropout": 0.1,
7
+ "bos_token_id": null,
8
+ "dim": 768,
9
+ "dropout": 0.1,
10
+ "eos_token_id": null,
11
+ "hidden_dim": 3072,
12
+ "initializer_range": 0.02,
13
+ "max_position_embeddings": 512,
14
+ "model_type": "distilbert",
15
+ "n_heads": 12,
16
+ "n_layers": 6,
17
+ "pad_token_id": 0,
18
+ "qa_dropout": 0.1,
19
+ "seq_classif_dropout": 0.2,
20
+ "sinusoidal_pos_embds": false,
21
+ "tie_weights_": true,
22
+ "tie_word_embeddings": true,
23
+ "transformers_version": "5.0.0",
24
+ "vocab_size": 30522
25
+ }
experiment_config.yaml ADDED
@@ -0,0 +1,131 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ project_name: security_nlp_3task
2
+ tasks:
3
+ - name: fraud_detection
4
+ type: single_label_classification
5
+ data_path: fixtures/sample_classification.jsonl
6
+ heads:
7
+ - name: fraud
8
+ num_labels: 2
9
+ weight: 1.0
10
+ dropout: 0.1
11
+ text_column: text
12
+ label_column: label
13
+ label_maps: null
14
+ - name: sentiment_analysis
15
+ type: single_label_classification
16
+ data_path: fixtures/sample_sentiment.jsonl
17
+ heads:
18
+ - name: sentiment
19
+ num_labels: 3
20
+ weight: 1.0
21
+ dropout: 0.1
22
+ text_column: text
23
+ label_column: label
24
+ label_maps: null
25
+ - name: call_quality
26
+ type: question_answering
27
+ data_path: fixtures/synthetic_qa_metrics_data_v01x.json
28
+ heads:
29
+ - name: opening
30
+ num_labels: 1
31
+ weight: 1.0
32
+ dropout: 0.1
33
+ - name: listening
34
+ num_labels: 5
35
+ weight: 1.5
36
+ dropout: 0.1
37
+ - name: proactiveness
38
+ num_labels: 3
39
+ weight: 1.0
40
+ dropout: 0.1
41
+ - name: resolution
42
+ num_labels: 5
43
+ weight: 2.0
44
+ dropout: 0.1
45
+ - name: hold
46
+ num_labels: 2
47
+ weight: 0.5
48
+ dropout: 0.1
49
+ - name: closing
50
+ num_labels: 1
51
+ weight: 1.0
52
+ dropout: 0.1
53
+ text_column: transcript
54
+ label_column: labels
55
+ label_maps:
56
+ opening:
57
+ 0: greeting
58
+ listening:
59
+ 0: acknowledgment
60
+ 1: empathy
61
+ 2: clarification
62
+ 3: active_listening
63
+ 4: patience
64
+ proactiveness:
65
+ 0: initiative
66
+ 1: follow_up
67
+ 2: suggestions
68
+ resolution:
69
+ 0: identified_issue
70
+ 1: provided_solution
71
+ 2: confirmed_resolution
72
+ 3: set_expectations
73
+ 4: offered_alternatives
74
+ hold:
75
+ 0: asked_permission
76
+ 1: explained_reason
77
+ closing:
78
+ 0: proper_farewell
79
+ model:
80
+ base_model: distilbert-base-uncased
81
+ hidden_size: 768
82
+ dropout: 0.1
83
+ fusion:
84
+ type: attention
85
+ dropout: 0.1
86
+ use_residual: true
87
+ num_attention_heads: 1
88
+ gate_init_value: 0.5
89
+ freeze_encoder_layers: 0
90
+ gradient_checkpointing: false
91
+ tokenizer:
92
+ max_length: 256
93
+ padding: max_length
94
+ truncation: true
95
+ training:
96
+ output_dir: ./results/3task_fusion
97
+ learning_rate: 2.0e-05
98
+ batch_size: 16
99
+ eval_batch_size: 16
100
+ num_epochs: 12
101
+ weight_decay: 0.01
102
+ warmup_steps: 20
103
+ max_grad_norm: 1.0
104
+ gradient_accumulation_steps: 1
105
+ use_amp: false
106
+ device: auto
107
+ task_sampling: proportional
108
+ temperature: 2.0
109
+ early_stopping_patience: 5
110
+ metric_for_best_model: eval_loss
111
+ greater_is_better: false
112
+ logging:
113
+ service: tensorboard
114
+ experiment_name: security_3task_fusion
115
+ tracking_uri: null
116
+ log_every_n_steps: 5
117
+ checkpoint:
118
+ save_every_n_epochs: 1
119
+ save_best: true
120
+ max_checkpoints: 2
121
+ data:
122
+ test_size: 0.2
123
+ seed: 42
124
+ num_workers: 0
125
+ pin_memory: true
126
+ pii_redaction:
127
+ enabled: false
128
+ strategy: mask
129
+ detect_types: null
130
+ hash_salt: jenga-ai-pii
131
+ log_detections: true
logs/security_3task_fusion/events.out.tfevents.1771060680.fb2f16b9fe6c.3306.0 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1e75987d635a183c6bfcdb36ea643fa4f995b5bf44184bc68d30e03baed9cd85
3
+ size 10389
logs/security_3task_fusion/events.out.tfevents.1771060870.fb2f16b9fe6c.4147.0 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5efa0186c0009692b6653c366bc16445f6059cbc30792568ca6b0e7db996e5a8
3
+ size 40069
metadata.json ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "base_model": "distilbert-base-uncased",
3
+ "hidden_size": 768,
4
+ "num_tasks": 3,
5
+ "task_names": [
6
+ "fraud_detection",
7
+ "sentiment_analysis",
8
+ "call_quality"
9
+ ],
10
+ "task_types": [
11
+ "single_label_classification",
12
+ "single_label_classification",
13
+ "question_answering"
14
+ ],
15
+ "has_fusion": true
16
+ }
model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3a06d1dc18f367e75599ea712ea481525e701f446c2660ba016f124e6f67746a
3
+ size 270300931