| --- |
| datasets: |
| - ner_dataset_2.jsonl |
| language: |
| - en |
| license: apache-2.0 |
| model-index: |
| - name: ner-distilbert-base-cased |
| results: |
| - dataset: |
| name: ner_dataset_2.jsonl |
| type: ner_dataset_2.jsonl |
| metrics: |
| - name: Eval Loss |
| type: eval_loss |
| value: 0.0216 |
| - name: Eval Accuracy |
| type: eval_accuracy |
| value: 0.993 |
| - name: Eval F1 |
| type: eval_f1 |
| value: 0.9929 |
| - name: Eval Recall |
| type: eval_recall |
| value: 0.993 |
| - name: Eval Precision |
| type: eval_precision |
| value: 0.9933 |
| task: |
| name: Ner |
| type: token-classification |
| tags: |
| - ner |
| - sklearn |
| - mlflow |
| - transformers |
| - openchs |
| --- |
| |
| # ner-distilbert-base-cased |
|
|
| This is a Named Entity Recognition (NER) model based on the `distilbert-base-cased` architecture. |
| This model was **expertly fine-tuned** to serve a critical function: accurately identifying and |
| classifying key information within the specific domain of child helpline conversations. |
| Through its specialized training, the model has been optimized to excel at recognizing the following |
| entities crucial for this context: |
| |
| `CALLER`: The person initiating the call for help. |
| `COUNSELOR`: The helpline staff member providing assistance. |
| `VICTIM`: The person who is the subject of the issue being discussed. |
| `PERPETRATOR`: The person causing the harm or issue. |
| `LOCATION`: Geographic places relevant to the conversation. |
| `AGE`: The age of the individuals involved. |
| `GENDER`: Gender references. |
| `INCIDENT_TYPE`: The specific type of problem or issue being reported (e.g., "bullying", "abuse" |
| ). |
| |
| ## Model Details |
|
|
| - **Model Name:** ner-distilbert-base-cased |
| - **Version:** 1 |
| - **Task:** Ner |
| - **Languages:** en |
| - **Framework:** sklearn |
| - **License:** apache-2.0 |
|
|
| ## Intended Uses & Limitations |
|
|
| ### Scope of Use |
| **In Scope:** The model is intended for text that is structurally and contextually similar to |
| the training data (i.e., conversational, first-person accounts of issues). |
| **Out of Scope:** |
| This model is **not intended for general-purpose NER tasks** (e.g., analyzing |
| news articles, legal documents, or emails). Its performance on text outside of its specialized |
| domain has not been evaluated. |
| |
| ## Training Data |
|
|
| - **Dataset:** ner_dataset_1.jsonl |
| - **Size:** Not specified |
| - **Languages:** en |
| - **Nature:** The training data is synthetic, generated to mimic real-world conversations while protecting the privacy and confidentiality of actual helpline users. |
|
|
| ## Training Configuration |
|
|
| | Parameter | Value | |
| |-----------|-------| |
| | Author | Rogendo | |
| | Batch Size | 4 | |
| | Epochs | 10 | |
| | Lr | 2e-05 | |
| | Model Name | distilbert-base-cased | |
| | Test Size | 0.1 | |
| | Training Date | 2025-10-30T11:58:48.315647 | |
| | Weight Decay | 0.01 | |
|
|
| ## Performance Metrics |
|
|
| ### Evaluation Results |
| | Metric | Value | |
| |--------|-------| |
| | Epoch | 10.0000 | |
| | Eval Accuracy | 0.9930 | |
| | Eval F1 | 0.9929 | |
| | Eval Loss | 0.0216 | |
| | Eval Precision | 0.9933 | |
| | Eval Recall | 0.9930 | |
| | Eval Runtime | 0.1509 | |
| | Eval Samples Per Second | 106.0170 | |
| | Eval Steps Per Second | 13.2520 | |
|
|
| ## Usage |
|
|
| ### Installation |
| ```bash |
| pip install transformers torch |
| ``` |
|
|
| ### Named Entity Recognition Example |
| ```python |
| from transformers import pipeline |
| |
| Use the model from the openchs repository on Hugging Face |
| ner = pipeline("ner", model="openchs/ner_distillbert_v1", aggregation_strategy="simple") |
| |
| Example of a typical helpline conversation snippet |
| text = "Hello 116, my name is Mary and I'm calling from Kampala. My daughter, Jane, is 12 years old and is being |
| bullied at school by a boy named Peter." |
| |
| entities = ner(text) |
| for entity in entities: |
| print(f"{entity['entity_group']}: {entity['word']} (score: {entity['score']:.2f})") |
| ``` |
|
|
|
|
| ## MLflow Tracking |
|
|
| - **Experiment:** NER_Distilbert/marlon |
| - **Run ID:** `10d2648a456a4f6ab74022a9e45c9f40` |
| - **Training Date:** 2025-10-30 11:58:48 |
| - **Tracking URI:** http://192.168.10.6:5000 |
| |
| ## Training Metrics Visualization |
| |
| View detailed training metrics and TensorBoard logs in the [Training metrics](https://huggingface.co/openchs/ner_distillbert_v1/tensorboard) tab. |
| |
| ## Citation |
| |
| ```bibtex |
| @misc{ner_distilbert_base_cased, |
| title={ner-distilbert-base-cased}, |
| author={OpenCHS Team}, |
| year={2025}, |
| publisher={Hugging Face}, |
| url={https://huggingface.co/openchs/ner_distillbert_v1} |
| } |
| ``` |
| |
| ## Contact |
| |
| info@bitz-itc.com |
| |
| --- |
| *Model card auto-generated from MLflow* |
| |
