| WaterSheep |
| Copyright 2026 Samrat Dutta <samratduttaofficial@gmail.com> |
|
|
| The WaterSheep code and model weights are licensed under the Apache License, |
| Version 2.0 (see LICENSE). |
|
|
| ------------------------------------------------------------------------------- |
| Base model |
| ------------------------------------------------------------------------------- |
| The WaterSheep model is fine-tuned from ModernBERT-base by Answer.AI and |
| LightOn (https://huggingface.co/answerdotai/ModernBERT-base), licensed under |
| the Apache License, Version 2.0. The encoder weights were modified by |
| fine-tuning, and a decision head was added. |
|
|
| ------------------------------------------------------------------------------- |
| Synthetic training data |
| ------------------------------------------------------------------------------- |
| Synthetic training examples, label descriptions and soft labels were produced |
| with Qwen3.5-4B by the Qwen team, Alibaba Cloud |
| (https://huggingface.co/Qwen/Qwen3.5-4B), licensed under the Apache License, |
| Version 2.0. |
|
|
| ------------------------------------------------------------------------------- |
| Public training data |
| ------------------------------------------------------------------------------- |
| The model was also trained on the public datasets below. The datasets are not |
| included in this repository or in the model files; the code downloads them |
| from their sources. Each dataset remains under its own license, as stated by |
| its source, and the Apache License of this project does not replace it. |
|
|
| Hugging Face (https://huggingface.co/datasets/<name>) |
| allenai/ai2_arc CC BY-SA 4.0 |
| allenai/openbookqa Apache 2.0 |
| allenai/prosocial-dialog CC BY 4.0 |
| allenai/quartz CC BY 4.0 |
| allenai/reward-bench ODC-By 1.0 |
| allenai/scitail Apache 2.0 |
| allenai/winogrande CC BY |
| Anthropic/hh-rlhf MIT |
| aps/super_glue (COPA) BSD 2-Clause |
| aps/super_glue (WSC) CC BY 4.0 |
| benayas/snips Apache 2.0 |
| bitext/Bitext-customer-support-llm-chatbot-training-dataset CDLA-Sharing 1.0 |
| cais/mmlu MIT |
| chengxuphd/liar2 Apache 2.0 |
| clinc/clinc_oos CC BY 3.0 |
| coastalcph/lex_glue CC BY 4.0 |
| deepset/prompt-injections Apache 2.0 |
| demelin/moral_stories MIT |
| Deysi/spam-detection-dataset Apache 2.0 |
| fancyzhx/dbpedia_14 CC BY-SA 3.0 |
| GBaker/MedQA-USMLE-4-options CC BY 4.0 |
| gfissore/arxiv-abstracts-2021 CC0 1.0 |
| glaiveai/glaive-function-calling-v2 Apache 2.0 |
| gonglinyuan/CoSQA MIT |
| google-research-datasets/go_emotions Apache 2.0 |
| google-research-datasets/paws Free for any purpose |
| google-research-datasets/poem_sentiment CC BY 4.0 |
| google/boolq CC BY-SA 3.0 |
| google/civil_comments CC0 1.0 |
| gretelai/symptom_to_diagnosis Apache 2.0 |
| hendrycks/ethics MIT |
| HuggingFaceH4/ultrafeedback_binarized MIT |
| Intel/orca_dpo_pairs Apache 2.0 |
| jackhhao/jailbreak-classification Apache 2.0 |
| jakartaresearch/semeval-absa CC BY 4.0 |
| lmsys/mt_bench_human_judgments CC BY 4.0 |
| marksverdhei/clickbait_title_classification MIT |
| mikex86/stackoverflow-posts CC BY-SA |
| mmathys/openai-moderation-api-evaluation MIT |
| mteb/amazon_counterfactual CC BY 4.0 |
| mteb/banking77 MIT |
| mteb/toxic_conversations_50k CC BY 4.0 |
| nvidia/Aegis-AI-Content-Safety-Dataset-2.0 CC BY 4.0 |
| nvidia/HelpSteer CC BY 4.0 |
| nvidia/HelpSteer2 CC BY 4.0 |
| nyu-mll/glue (QNLI) CC BY-SA 4.0 |
| nyu-mll/multi_nli OANC / CC BY-SA 3.0 / CC BY 3.0 |
| openlifescienceai/medmcqa Apache 2.0 |
| owaiskha9654/PubMed_MultiLabel_Text_Classification_Dataset_MeSH AFL 3.0 |
| pminervini/HaluEval Apache 2.0 |
| prometheus-eval/Feedback-Collection CC BY 4.0 |
| prometheus-eval/Preference-Collection CC BY 4.0 |
| qiaojin/PubMedQA MIT |
| rajpurkar/squad_v2 CC BY-SA 4.0 |
| reshabhs/SPML_Chatbot_Prompt_Injection MIT |
| SetFit/amazon_massive_intent_en-US CC BY 4.0 |
| SetFit/amazon_massive_scenario_en-US CC BY 4.0 |
| SetFit/student-question-categories CC0 1.0 |
| stanfordnlp/snli CC BY-SA 4.0 |
| tals/vitaminc CC BY-SA 3.0 |
| tasksource/bigbench Apache 2.0 |
| tasksource/crowdflower (political media subsets) CC0 1.0 |
| tasksource/esci Apache 2.0 |
| tasksource/folio CC BY-SA 4.0 |
| tau/commonsense_qa MIT |
| tdavidson/hate_speech_offensive MIT |
| thesofakillers/jigsaw-toxic-comment-classification-challenge CC BY-SA 3.0 |
| TIGER-Lab/MMLU-Pro MIT |
| TimSchopf/medical_abstracts CC BY-SA 3.0 |
| truthfulqa/truthful_qa Apache 2.0 |
| ucirvine/sms_spam CC BY 4.0 |
| zeroshot/twitter-financial-news-sentiment MIT |
| zeroshot/twitter-financial-news-topic MIT |
|
|
| Kaggle (https://www.kaggle.com/datasets/<name>) |
| andrewmvd/cyberbullying-classification CC BY 4.0 |
| imoore/60k-stack-overflow-questions-with-quality-rate MIT / CC BY-SA |
| jp797498e/twitter-entity-sentiment-analysis CC0 1.0 |
| nicapotato/womens-ecommerce-clothing-reviews CC0 1.0 |
| rmisra/clothing-fit-dataset-for-size-recommendation CC BY 4.0 |
| rounakbanik/the-movies-dataset CC0 1.0 |
| saurabhshahane/ecommerce-text-classification CC BY 4.0 |
| shivamb/real-or-fake-fake-jobposting-prediction CC0 1.0 |
| snap/amazon-fine-food-reviews CC0 1.0 |
| snehaanbhawal/resume-dataset CC0 1.0 |
| subhajournal/phishingemails LGPL 3.0 |
| tboyle10/medicaltranscriptions CC0 1.0 |
|
|
| Other sources |
| NLU Evaluation Data (HWU64) CC BY 4.0 |
| https://github.com/xliuhw/NLU-Evaluation-Data |
| UCI News Aggregator CC BY 4.0 |
| https://archive.ics.uci.edu/dataset/359/news+aggregator |
| UCI YouTube Spam Collection CC BY 4.0 |
| https://archive.ics.uci.edu/dataset/380/youtube+spam+collection |
|
|
| ------------------------------------------------------------------------------- |
| Held-out data |
| ------------------------------------------------------------------------------- |
| The pipeline also downloads these datasets but withholds them from training; |
| they are used only to test the model on data it has not seen. |
|
|
| allenai/qasc CC BY 4.0 |
| https://huggingface.co/datasets/allenai/qasc |
| PromptCloudHQ/amazon-reviews-unlocked-mobile-phones CC0 1.0 |
| https://www.kaggle.com/datasets/PromptCloudHQ/amazon-reviews-unlocked-mobile-phones |
| rmisra/news-category-dataset CC BY 4.0 |
| https://www.kaggle.com/datasets/rmisra/news-category-dataset |
| rmisra/news-headlines-dataset-for-sarcasm-detection CC BY 4.0 |
| https://www.kaggle.com/datasets/rmisra/news-headlines-dataset-for-sarcasm-detection |
|
|
