datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
pushbackchat-v1
PushbackChat
PushbackChat is a curated dataset of real-world user pushback events extracted
from WildChat-4.8M. The v1 release contains 2,309 English pushback events:
factual_correction: 750
clarification_demand: 750
preference_disagreement: 750
emotional_pushback: 59
This anonymous release contains redacted copies of the primary reproducibility
artifacts named in the paper appendix:
dataset/pushbackchat_3k_v1.json
dataset/pushbackchat_3k_v1.json.sha256
config/default.yaml… See the full description on the dataset page: https://huggingface.co/datasets/NoirZangetsu/pushbackchat-v1.bankless_ROLLUP_BTC_1T_Market_Cap__More_ETH_ETF_Filings__STRK_Airdrop_Pushbackpush_backward_compatibility
Dataset Card for "push_backward_compatibility"
More Information needed
Gentle-Pushback-v2-shareGPTGentle-Pushback-8.5k-alpaca
Gentle Pushback
This data set was created to limit sycofancy in language models and encouraging the models to (gently) push back and call out bad ideas.
Samples from other data sets were mixed and the responses rewritten with that in mind. Each entry was rewritten twice and scored on a number of factors, keeping only the highest scoring response to each entry.