Upload JengaAI model: Police_Case_Multi-Task_v2 — Mar 09 00:25

README.md

@@ -1,3 +1,41 @@
----
-license: gpl
----
+---
+tags:
+  - jenga-ai
+  - nlp
+pipeline_tag: text-classification
+metrics:
+  - accuracy
+  - f1
+---
+# Police_Case_Multi-Task_v2 — Mar 09 00:25
+
+Fine-tuned with [JengaAI](https://github.com/TujengeAI/JengaAI) — a multi-task NLP training framework.
+
+| Metric | Value |
+|--------|-------|
+| Accuracy | N/A |
+| F1 Score | N/A |
+| Modality | NLP |
+
+## Loading with JengaAI SDK
+
+This model uses a custom `MultiTaskModel` architecture and **requires the JengaAI SDK** to run inference.
+
+```python
+from jenga_ai.inference.predictor import MultiTaskPredictor
+
+predictor = MultiTaskPredictor.from_checkpoint(
+    model_dir="Rogendo/police_case_reporting_multitask_distilbert",   # HuggingFace repo ID
+)
+result = predictor.predict("Your text here")
+print(result)
+```
+
+## Artifacts
+
+| File | Description |
+|------|-------------|
+| `model.pt` | Trained model weights (PyTorch) |
+| `experiment_config.yaml` | Full training config with label maps |
+| `label_maps.json` | Class label mappings per task |
+| `encoder_config/` | Base encoder tokenizer & config |

checkpoints/best/checkpoint_meta.json

@@ -0,0 +1,32 @@
+{
+  "epoch": 0,
+  "metrics": {
+    "sentiment_sentiment_accuracy": 0.5,
+    "sentiment_sentiment_precision": 0.5,
+    "sentiment_sentiment_recall": 0.5,
+    "sentiment_sentiment_f1": 0.5,
+    "main_category_main_category_accuracy": 0.0,
+    "main_category_main_category_precision": 0.0,
+    "main_category_main_category_recall": 0.0,
+    "main_category_main_category_f1": 0.0,
+    "subcategory_subcategory_accuracy": 0.0,
+    "subcategory_subcategory_precision": 0.0,
+    "subcategory_subcategory_recall": 0.0,
+    "subcategory_subcategory_f1": 0.0,
+    "priority_priority_accuracy": 0.0,
+    "priority_priority_precision": 0.0,
+    "priority_priority_recall": 0.0,
+    "priority_priority_f1": 0.0,
+    "intervention_intervention_accuracy": 0.0,
+    "intervention_intervention_precision": 0.0,
+    "intervention_intervention_recall": 0.0,
+    "intervention_intervention_f1": 0.0,
+    "entity_extraction_entity_extraction_accuracy": 0.7372881355932204,
+    "entity_extraction_entity_extraction_precision": 0.7372881355932204,
+    "entity_extraction_entity_extraction_recall": 0.7372881355932204,
+    "entity_extraction_entity_extraction_f1": 0.7372881355932204,
+    "eval_loss": 2.293487270673116,
+    "train_loss_avg": 2.2960834205150604
+  },
+  "is_best": true
+}

checkpoints/best/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ef8b20ff817c8f14541763fd58b45d1cc75b90f658a0c8dcf8640e4e7e7c6dff
+size 270439507

checkpoints/epoch_2/checkpoint_meta.json

@@ -0,0 +1,32 @@
+{
+  "epoch": 2,
+  "metrics": {
+    "sentiment_sentiment_accuracy": 0.5,
+    "sentiment_sentiment_precision": 0.5,
+    "sentiment_sentiment_recall": 0.5,
+    "sentiment_sentiment_f1": 0.5,
+    "main_category_main_category_accuracy": 0.0,
+    "main_category_main_category_precision": 0.0,
+    "main_category_main_category_recall": 0.0,
+    "main_category_main_category_f1": 0.0,
+    "subcategory_subcategory_accuracy": 0.0,
+    "subcategory_subcategory_precision": 0.0,
+    "subcategory_subcategory_recall": 0.0,
+    "subcategory_subcategory_f1": 0.0,
+    "priority_priority_accuracy": 0.0,
+    "priority_priority_precision": 0.0,
+    "priority_priority_recall": 0.0,
+    "priority_priority_f1": 0.0,
+    "intervention_intervention_accuracy": 0.0,
+    "intervention_intervention_precision": 0.0,
+    "intervention_intervention_recall": 0.0,
+    "intervention_intervention_f1": 0.0,
+    "entity_extraction_entity_extraction_accuracy": 0.847457627118644,
+    "entity_extraction_entity_extraction_precision": 0.7181844297615627,
+    "entity_extraction_entity_extraction_recall": 0.847457627118644,
+    "entity_extraction_entity_extraction_f1": 0.7774840615767377,
+    "eval_loss": 2.5300551454226174,
+    "train_loss_avg": 1.4021934138403997
+  },
+  "is_best": false
+}

checkpoints/epoch_2/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6fa46fc88dee4fdefa84f807dd62c8e458004e2c041f5616f271f9750a761a74
+size 270439507

checkpoints/epoch_3/checkpoint_meta.json

@@ -0,0 +1,32 @@
+{
+  "epoch": 3,
+  "metrics": {
+    "sentiment_sentiment_accuracy": 0.5,
+    "sentiment_sentiment_precision": 0.5,
+    "sentiment_sentiment_recall": 0.5,
+    "sentiment_sentiment_f1": 0.5,
+    "main_category_main_category_accuracy": 0.0,
+    "main_category_main_category_precision": 0.0,
+    "main_category_main_category_recall": 0.0,
+    "main_category_main_category_f1": 0.0,
+    "subcategory_subcategory_accuracy": 0.0,
+    "subcategory_subcategory_precision": 0.0,
+    "subcategory_subcategory_recall": 0.0,
+    "subcategory_subcategory_f1": 0.0,
+    "priority_priority_accuracy": 0.0,
+    "priority_priority_precision": 0.0,
+    "priority_priority_recall": 0.0,
+    "priority_priority_f1": 0.0,
+    "intervention_intervention_accuracy": 0.0,
+    "intervention_intervention_precision": 0.0,
+    "intervention_intervention_recall": 0.0,
+    "intervention_intervention_f1": 0.0,
+    "entity_extraction_entity_extraction_accuracy": 0.847457627118644,
+    "entity_extraction_entity_extraction_precision": 0.7181844297615627,
+    "entity_extraction_entity_extraction_recall": 0.847457627118644,
+    "entity_extraction_entity_extraction_f1": 0.7774840615767377,
+    "eval_loss": 2.5224763651688895,
+    "train_loss_avg": 1.2216199677851465
+  },
+  "is_best": false
+}

checkpoints/epoch_3/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a29701fc9e837a557fd5b08f640ce9bff8876f6e34a933e4614346dfd4b38d4d
+size 270439507

config.yaml

@@ -0,0 +1,112 @@
+project_name: police_case_multitask
+
+model:
+  base_model: distilbert-base-uncased
+  dropout: 0.1
+  freeze_encoder_layers: 0
+  gradient_checkpointing: false
+  fusion:
+    type: attention
+    dropout: 0.1
+    use_residual: true
+    num_attention_heads: 1
+
+tokenizer:
+  max_length: 512
+  padding: max_length
+  truncation: true
+
+training:
+  output_dir: ./results/police_case_multitask
+  learning_rate: 0.00002
+  batch_size: 4
+  num_epochs: 3
+  weight_decay: 0.01
+  warmup_steps: 50
+  max_grad_norm: 1.0
+  gradient_accumulation_steps: 1
+  use_amp: false
+  device: auto
+  early_stopping_patience: 3
+  metric_for_best_model: eval_loss
+  greater_is_better: false
+  logging:
+    service: mlflow
+    experiment_name: police_case_multitask
+    tracking_uri: http://localhost:5000
+    log_every_n_steps: 10
+  checkpoint:
+    save_every_n_epochs: 1
+    save_best: true
+    max_checkpoints: 2
+  data:
+    test_size: 0.15
+    seed: 42
+    num_workers: 0
+
+tasks:
+  - name: sentiment
+    type: sentiment
+    data_path: test_data/police_cases/police_case_reports.json
+    text_column: report_text
+    label_column: sentiment
+    heads:
+      - name: sentiment
+        num_labels: 4
+        weight: 1.0
+        dropout: 0.1
+
+  - name: main_category
+    type: single_label_classification
+    data_path: test_data/police_cases/police_case_reports.json
+    text_column: report_text
+    label_column: main_category
+    heads:
+      - name: main_category
+        num_labels: 10
+        weight: 1.0
+        dropout: 0.1
+
+  - name: subcategory
+    type: single_label_classification
+    data_path: test_data/police_cases/police_case_reports.json
+    text_column: report_text
+    label_column: subcategory
+    heads:
+      - name: subcategory
+        num_labels: 12
+        weight: 1.0
+        dropout: 0.1
+
+  - name: priority
+    type: single_label_classification
+    data_path: test_data/police_cases/police_case_reports.json
+    text_column: report_text
+    label_column: priority
+    heads:
+      - name: priority
+        num_labels: 4
+        weight: 1.0
+        dropout: 0.1
+
+  - name: intervention
+    type: single_label_classification
+    data_path: test_data/police_cases/police_case_reports.json
+    text_column: report_text
+    label_column: intervention
+    heads:
+      - name: intervention
+        num_labels: 7
+        weight: 1.0
+        dropout: 0.1
+
+  - name: entity_extraction
+    type: ner
+    data_path: test_data/police_cases/police_case_reports.json
+    text_column: report_text
+    label_column: entities
+    heads:
+      - name: entity_extraction
+        num_labels: 15
+        weight: 1.0
+        dropout: 0.1

encoder_config/config.json

@@ -0,0 +1,25 @@
+{
+  "activation": "gelu",
+  "architectures": [
+    "DistilBertForMaskedLM"
+  ],
+  "attention_dropout": 0.1,
+  "bos_token_id": null,
+  "dim": 768,
+  "dropout": 0.1,
+  "eos_token_id": null,
+  "hidden_dim": 3072,
+  "initializer_range": 0.02,
+  "max_position_embeddings": 512,
+  "model_type": "distilbert",
+  "n_heads": 12,
+  "n_layers": 6,
+  "pad_token_id": 0,
+  "qa_dropout": 0.1,
+  "seq_classif_dropout": 0.2,
+  "sinusoidal_pos_embds": false,
+  "tie_weights_": true,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.1.0",
+  "vocab_size": 30522
+}

experiment_config.yaml

@@ -0,0 +1,196 @@
+project_name: police_case_reports_json
+tasks:
+- name: sentiment
+  type: sentiment
+  data_path: /home/naynek/Desktop/TujengeAI/Update/JengaAI/backend/uploads/datasets/16ced0930e7e_police_case_reports.json
+  heads:
+  - name: sentiment
+    num_labels: 4
+    weight: 1.0
+    dropout: 0.1
+  text_column: report_text
+  label_column: sentiment
+  label_maps:
+    sentiment:
+      0: negative
+      1: neutral
+      2: positive
+      3: urgent
+- name: main_category
+  type: single_label_classification
+  data_path: /home/naynek/Desktop/TujengeAI/Update/JengaAI/backend/uploads/datasets/16ced0930e7e_police_case_reports.json
+  heads:
+  - name: main_category
+    num_labels: 10
+    weight: 1.0
+    dropout: 0.1
+  text_column: report_text
+  label_column: main_category
+  label_maps:
+    main_category:
+      0: Child Abuse
+      1: Commendation
+      2: Drug Abuse
+      3: Fraud
+      4: Human Trafficking
+      5: Missing Person
+      6: Theft
+      7: Traffic
+      8: Vandalism
+      9: Violence
+- name: subcategory
+  type: single_label_classification
+  data_path: /home/naynek/Desktop/TujengeAI/Update/JengaAI/backend/uploads/datasets/16ced0930e7e_police_case_reports.json
+  heads:
+  - name: subcategory
+    num_labels: 24
+    weight: 1.0
+    dropout: 0.1
+  text_column: report_text
+  label_column: subcategory
+  label_maps:
+    subcategory:
+      0: ATM Fraud
+      1: Abandoned Child
+      2: Abandoned Vehicle
+      3: Bank Robbery
+      4: Breaking and Entering
+      5: Child Labour
+      6: Consumer Fraud
+      7: Domestic Servitude
+      8: Domestic Violence
+      9: Drug Distribution
+      10: Extortion
+      11: Gunfire
+      12: Hit and Run
+      13: Identity Theft
+      14: Illicit Brewing
+      15: Missing Child
+      16: Mugging
+      17: Officer Commendation
+      18: Phone Scam
+      19: Property Destruction
+      20: Robbery
+      21: Sexual Abuse
+      22: Tourist Scam
+      23: Vehicle Theft
+- name: priority
+  type: single_label_classification
+  data_path: /home/naynek/Desktop/TujengeAI/Update/JengaAI/backend/uploads/datasets/16ced0930e7e_police_case_reports.json
+  heads:
+  - name: priority
+    num_labels: 4
+    weight: 1.0
+    dropout: 0.1
+  text_column: report_text
+  label_column: priority
+  label_maps:
+    priority:
+      0: critical
+      1: high
+      2: low
+      3: medium
+- name: intervention
+  type: single_label_classification
+  data_path: /home/naynek/Desktop/TujengeAI/Update/JengaAI/backend/uploads/datasets/16ced0930e7e_police_case_reports.json
+  heads:
+  - name: intervention
+    num_labels: 8
+    weight: 1.0
+    dropout: 0.1
+  text_column: report_text
+  label_column: intervention
+  label_maps:
+    intervention:
+      0: Counseling
+      1: Investigation
+      2: Legal Aid
+      3: Medical
+      4: None Required
+      5: Police Dispatch
+      6: Search and Rescue
+      7: Welfare Services
+- name: entity_extraction
+  type: ner
+  data_path: /home/naynek/Desktop/TujengeAI/Update/JengaAI/backend/uploads/datasets/16ced0930e7e_police_case_reports.json
+  heads:
+  - name: entity_extraction
+    num_labels: 15
+    weight: 1.0
+    dropout: 0.1
+  text_column: report_text
+  label_column: entities
+  label_maps:
+    ner_head:
+      0: O
+      1: B-DATE
+      2: I-DATE
+      3: B-LOCATION
+      4: I-LOCATION
+      5: B-NAME
+      6: I-NAME
+      7: B-ORG
+      8: I-ORG
+      9: B-TIME
+      10: I-TIME
+      11: B-VEHICLE
+      12: I-VEHICLE
+      13: B-WEAPON
+      14: I-WEAPON
+model:
+  base_model: distilbert-base-uncased
+  hidden_size: 768
+  dropout: 0.1
+  fusion:
+    type: attention
+    dropout: 0.1
+    use_residual: true
+    num_attention_heads: 1
+    gate_init_value: 0.5
+  freeze_encoder_layers: 0
+  gradient_checkpointing: false
+tokenizer:
+  max_length: 512
+  padding: max_length
+  truncation: true
+training:
+  output_dir: ./results/police_case_reports_json
+  learning_rate: 2.0e-05
+  batch_size: 4
+  eval_batch_size: 4
+  num_epochs: 4
+  weight_decay: 0.01
+  warmup_steps: 50
+  max_grad_norm: 1.0
+  gradient_accumulation_steps: 1
+  use_amp: false
+  device: auto
+  task_sampling: round_robin
+  temperature: 2.0
+  early_stopping_patience: 3
+  metric_for_best_model: eval_loss
+  greater_is_better: false
+  logging:
+    service: mlflow
+    experiment_name: police_case_reports_json
+    tracking_uri: http://localhost:5000
+    log_every_n_steps: 10
+  checkpoint:
+    save_every_n_epochs: 1
+    save_best: true
+    max_checkpoints: 2
+  data:
+    test_size: 0.08
+    seed: 42
+    num_workers: 0
+    pin_memory: true
+    pii_redaction:
+      enabled: false
+      detect_types:
+      - email
+      - phone
+      - url
+      - card
+      - name
+      - org
+      log_detections: true

label_maps.json

@@ -0,0 +1,91 @@
+{
+  "sentiment": {
+    "sentiment": {
+      "0": "negative",
+      "1": "neutral",
+      "2": "positive",
+      "3": "urgent"
+    }
+  },
+  "main_category": {
+    "main_category": {
+      "0": "Child Abuse",
+      "1": "Commendation",
+      "2": "Drug Abuse",
+      "3": "Fraud",
+      "4": "Human Trafficking",
+      "5": "Missing Person",
+      "6": "Theft",
+      "7": "Traffic",
+      "8": "Vandalism",
+      "9": "Violence"
+    }
+  },
+  "subcategory": {
+    "subcategory": {
+      "0": "ATM Fraud",
+      "1": "Abandoned Child",
+      "2": "Abandoned Vehicle",
+      "3": "Bank Robbery",
+      "4": "Breaking and Entering",
+      "5": "Child Labour",
+      "6": "Consumer Fraud",
+      "7": "Domestic Servitude",
+      "8": "Domestic Violence",
+      "9": "Drug Distribution",
+      "10": "Extortion",
+      "11": "Gunfire",
+      "12": "Hit and Run",
+      "13": "Identity Theft",
+      "14": "Illicit Brewing",
+      "15": "Missing Child",
+      "16": "Mugging",
+      "17": "Officer Commendation",
+      "18": "Phone Scam",
+      "19": "Property Destruction",
+      "20": "Robbery",
+      "21": "Sexual Abuse",
+      "22": "Tourist Scam",
+      "23": "Vehicle Theft"
+    }
+  },
+  "priority": {
+    "priority": {
+      "0": "critical",
+      "1": "high",
+      "2": "low",
+      "3": "medium"
+    }
+  },
+  "intervention": {
+    "intervention": {
+      "0": "Counseling",
+      "1": "Investigation",
+      "2": "Legal Aid",
+      "3": "Medical",
+      "4": "None Required",
+      "5": "Police Dispatch",
+      "6": "Search and Rescue",
+      "7": "Welfare Services"
+    }
+  },
+  "entity_extraction": {
+    "ner_head": {
+      "0": "O",
+      "1": "B-DATE",
+      "2": "I-DATE",
+      "3": "B-LOCATION",
+      "4": "I-LOCATION",
+      "5": "B-NAME",
+      "6": "I-NAME",
+      "7": "B-ORG",
+      "8": "I-ORG",
+      "9": "B-TIME",
+      "10": "I-TIME",
+      "11": "B-VEHICLE",
+      "12": "I-VEHICLE",
+      "13": "B-WEAPON",
+      "14": "I-WEAPON"
+    }
+  }
+}

metadata.json

@@ -0,0 +1,22 @@
+{
+  "base_model": "distilbert-base-uncased",
+  "hidden_size": 768,
+  "num_tasks": 6,
+  "task_names": [
+    "sentiment",
+    "main_category",
+    "subcategory",
+    "priority",
+    "intervention",
+    "entity_extraction"
+  ],
+  "task_types": [
+    "sentiment",
+    "single_label_classification",
+    "single_label_classification",
+    "single_label_classification",
+    "single_label_classification",
+    "ner"
+  ],
+  "has_fusion": true
+}

model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a29701fc9e837a557fd5b08f640ce9bff8876f6e34a933e4614346dfd4b38d4d
+size 270439507