Add CROWDLAB for LLM Evaluation notebook (#91)

Co-authored-by: Jonas Mueller <[email protected]>

README.md

@@ -20,17 +20,18 @@ To quickly learn how to run cleanlab on your own data, first check out the [quic
 | 10   | [huggingface_keras_imdb](huggingface_keras_imdb/huggingface_keras_imdb.ipynb)                                             |  CleanLearning for text classification with Keras Model + pretrained BERT backbone and Tensorflow Dataset.         |
 | 11   | [fasttext_amazon_reviews](fasttext_amazon_reviews/fasttext_amazon_reviews.ipynb)                         | Finding label errors in Amazon Reviews text dataset using a cleanlab-compatible [FastText model](https://github.com/cleanlab/cleanlab/blob/master/cleanlab/models/fasttext.py).                                                                                                    |
 | 12   | [multiannotator_cifar10](multiannotator_cifar10/multiannotator_cifar10.ipynb)                                             | Iteratively improve consensus labels and trained classifier from data labeled by multiple annotators.                                                            |
-| 13  | [active_learning_multiannotator](active_learning_multiannotator/active_learning.ipynb)                                             | Improve a classifier model by iteratively collecting additional labels from data annotators. This active learning pipeline considers data labeled in batches by multiple (imperfect) annotators.                                                             |
-| 14  | [active_learning_single_annotator](active_learning_single_annotator/active_learning_single_annotator.ipynb)                                             | Improve a classifier model by iteratively labeling batches of currently-unlabeled data.  This demonstrates a standard active learning pipeline with *at most one label* collected for each example (unlike our multi-annotator active learning notebook which allows re-labeling).                                                            |
-| 15  | [active_learning_transformers](active_learning_transformers/active_learning.ipynb)                                             | Improve a Transformer model for classifying politeness of text by iteratively labeling and re-labeling batches of data using multiple annotators.  If you haven't done active learning with re-labeling, try the [active_learning_multiannotator](active_learning_multiannotator/active_learning.ipynb) notebook first.                                          |
-| 16  | [outlier_detection_cifar10](outlier_detection_cifar10/outlier_detection_cifar10.ipynb)                                             | Train AutoML for image classification and use it to detect out-of-distribution images.                                                                                                 |
-| 17  | [multilabel_classification](multilabel_classification/image_tagging.ipynb)                                               | Find label errors in an image tagging dataset ([CelebA](https://mmlab.ie.cuhk.edu.hk/projects/CelebA.html)) using a [Pytorch model](multilabel_classification/pytorch_network_training.ipynb) you can easily train for multi-label classification. |
-| 18  | [entity_recognition](entity_recognition/)                                             | Train Transformer model  for Named Entity Recognition and produce out-of-sample `pred_probs` for **cleanlab.token_classification**.      |
-| 19  | [transformer_sklearn](transformer_sklearn/transformer_sklearn.ipynb)                                             | How to use `KerasWrapperModel` to make any Keras model sklearn-compatible, demonstrated here for a BERT Transformer.      |
-| 20  | [cnn_coteaching_cifar10](cnn_coteaching_cifar10/README.md)                                               | Train a [Convolutional Neural Network](https://github.com/cleanlab/cleanlab/blob/master/cleanlab/experimental/cifar_cnn.py) on noisily labeled Cifar10 image data using cleanlab with [coteaching](https://github.com/cleanlab/cleanlab/blob/master/cleanlab/experimental/coteaching.py).  |
-| 21  | [non_iid_detection](non_iid_detection/non_iid_detection.ipynb)  | Use Datalab to detect non-IID sampling (e.g. drift) in datasets based on numeric features or embeddings. |
-| 22  | [object_detection](object_detection/README.md)  | Train Detectron2 object detection model for use with cleanlab. |
-| 23  | [semantic segmentation](segmentation/training_ResNeXt50_for_Semantic_Segmentation_on_SYNTHIA.ipynb)  | Train ResNeXt semantic segmentation model for use with cleanlab. |
+| 13 | [llm_evals_w_crowdlab](llm_evals_w_crowdlab/llm_evals_w_crowdlab.ipynb) | Reliable LLM Evaluation with multiple human/AI reviewers of varying competency (via CROWDLAB and LLM-as-judge GPT token probabilities).                                                                                |
+| 14  | [active_learning_multiannotator](active_learning_multiannotator/active_learning.ipynb)                                             | Improve a classifier model by iteratively collecting additional labels from data annotators. This active learning pipeline considers data labeled in batches by multiple (imperfect) annotators.                                                             |
+| 15  | [active_learning_single_annotator](active_learning_single_annotator/active_learning_single_annotator.ipynb)                                             | Improve a classifier model by iteratively labeling batches of currently-unlabeled data.  This demonstrates a standard active learning pipeline with *at most one label* collected for each example (unlike our multi-annotator active learning notebook which allows re-labeling).                                                            |
+| 16  | [active_learning_transformers](active_learning_transformers/active_learning.ipynb)                                             | Improve a Transformer model for classifying politeness of text by iteratively labeling and re-labeling batches of data using multiple annotators.  If you haven't done active learning with re-labeling, try the [active_learning_multiannotator](active_learning_multiannotator/active_learning.ipynb) notebook first.                                          |
+| 17  | [outlier_detection_cifar10](outlier_detection_cifar10/outlier_detection_cifar10.ipynb)                                             | Train AutoML for image classification and use it to detect out-of-distribution images.                                                                                                 |
+| 18  | [multilabel_classification](multilabel_classification/image_tagging.ipynb)                                               | Find label errors in an image tagging dataset ([CelebA](https://mmlab.ie.cuhk.edu.hk/projects/CelebA.html)) using a [Pytorch model](multilabel_classification/pytorch_network_training.ipynb) you can easily train for multi-label classification. |
+| 19  | [entity_recognition](entity_recognition/)                                             | Train Transformer model  for Named Entity Recognition and produce out-of-sample `pred_probs` for **cleanlab.token_classification**.      |
+| 20  | [transformer_sklearn](transformer_sklearn/transformer_sklearn.ipynb)                                             | How to use `KerasWrapperModel` to make any Keras model sklearn-compatible, demonstrated here for a BERT Transformer.      |
+| 21  | [cnn_coteaching_cifar10](cnn_coteaching_cifar10/README.md)                                               | Train a [Convolutional Neural Network](https://github.com/cleanlab/cleanlab/blob/master/cleanlab/experimental/cifar_cnn.py) on noisily labeled Cifar10 image data using cleanlab with [coteaching](https://github.com/cleanlab/cleanlab/blob/master/cleanlab/experimental/coteaching.py).  |
+| 22  | [non_iid_detection](non_iid_detection/non_iid_detection.ipynb)  | Use Datalab to detect non-IID sampling (e.g. drift) in datasets based on numeric features or embeddings. |
+| 23  | [object_detection](object_detection/README.md)  | Train Detectron2 object detection model for use with cleanlab. |
+| 24  | [semantic segmentation](segmentation/training_ResNeXt50_for_Semantic_Segmentation_on_SYNTHIA.ipynb)  | Train ResNeXt semantic segmentation model for use with cleanlab. |
 
 
 ## Instructions