import os from typing import Dict def get_readme(model_name: str, metric: Dict, metric_span: Dict, config: Dict): language_model = config['model'] dataset = None dataset_alias = "custom" if config["dataset"] is not None: dataset = sorted([i for i in config["dataset"]]) dataset_alias = ','.join(dataset) config_text = "\n".join([f" - {k}: {v}" for k, v in config.items()]) ci_micro = '\n'.join([f' - {k}%: {v}' for k, v in metric["micro/f1_ci"].items()]) ci_macro = '\n'.join([f' - {k}%: {v}' for k, v in metric["micro/f1_ci"].items()]) per_entity_metric = '\n'.join([f'- {k}: {v["f1"]}' for k, v in metric['per_entity_metric'].items()]) if dataset is None: dataset_link = 'custom' else: dataset = [dataset] if type(dataset) is str else dataset dataset_link = ','.join([f"[{d}](https://huggingface.co/datasets/{d})" for d in dataset]) return f"""--- datasets: - {dataset_alias} metrics: - f1 - precision - recall model-index: - name: {model_name} results: - task: name: Token Classification type: token-classification dataset: name: {dataset_alias} type: {dataset_alias} args: {dataset_alias} metrics: - name: F1 type: f1 value: {metric['micro/f1']} - name: Precision type: precision value: {metric['micro/precision']} - name: Recall type: recall value: {metric['micro/recall']} - name: F1 (macro) type: f1_macro value: {metric['macro/f1']} - name: Precision (macro) type: precision_macro value: {metric['macro/precision']} - name: Recall (macro) type: recall_macro value: {metric['macro/recall']} - name: F1 (entity span) type: f1_entity_span value: {metric_span['micro/f1']} - name: Precision (entity span) type: precision_entity_span value: {metric_span['micro/precision']} - name: Recall (entity span) type: recall_entity_span value: {metric_span['micro/recall']} pipeline_tag: token-classification widget: - text: "Jacob Collier is a Grammy awarded artist from England." example_title: "NER Example 1" --- # {model_name} This model is a fine-tuned version of [{language_model}](https://huggingface.co/{language_model}) on the {dataset_link} dataset. Model fine-tuning is done via [T-NER](https://github.com/asahi417/tner)'s hyper-parameter search (see the repository for more detail). It achieves the following results on the test set: - F1 (micro): {metric['micro/f1']} - Precision (micro): {metric['micro/precision']} - Recall (micro): {metric['micro/recall']} - F1 (macro): {metric['macro/f1']} - Precision (macro): {metric['macro/precision']} - Recall (macro): {metric['macro/recall']} The per-entity breakdown of the F1 score on the test set are below: {per_entity_metric} For F1 scores, the confidence interval is obtained by bootstrap as below: - F1 (micro): {ci_micro} - F1 (macro): {ci_macro} Full evaluation can be found at [metric file of NER](https://huggingface.co/{model_name}/raw/main/eval/metric.json) and [metric file of entity span](https://huggingface.co/{model_name}/raw/main/eval/metric_span.json). ### Usage This model can be used through the [tner library](https://github.com/asahi417/tner). Install the library via pip ```shell pip install tner ``` and activate model as below. ```python from tner import TransformersNER model = TransformersNER("{model_name}") model.predict(["Jacob Collier is a Grammy awarded English artist from London"]) ``` It can be used via transformers library but it is not recommended as CRF layer is not supported at the moment. ### Training hyperparameters The following hyperparameters were used during training: {config_text} The full configuration can be found at [fine-tuning parameter file](https://huggingface.co/{model_name}/raw/main/trainer_config.json). ### Reference If you use any resource from T-NER, please consider to cite our [paper](https://aclanthology.org/2021.eacl-demos.7/). ``` {bib} ``` """