| 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 | | |