datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
8b_honesty_sft_10k70b_honesty_sft_10k70b_honesty_sft_10k_correct8b_honesty_sft_10k_correct8b_honesty_sft_10k_v2_majority8b-8925_honesty_sft_10k_0.2_0.5_0.8_correctness8b_honesty_sft_10k_v2_majority_correct8b_honesty_sft_10k_v1_majority8b_honesty_sft_10k_v1_majority_correct8b-8925_honesty_sft_10k_0.2_0.5_0.88b_honesty_sft_10k_v1_correctnesssurrogate-2-honesty-sft
axentx/surrogate-2-honesty-sft
Honesty / abstention / anti-hallucination SFT for surrogate-2.1 (3-5% in-mix, esp biz/trader/devmode + ALL abstention axis). Slice when2call_abstention: nvidia/When2Call SFT (CC-BY-4.0, 15k) — when NOT to call a tool / ask-followup / admit-can't-answer, kept as both {prompt,response} and {tools,messages}. Slice when2call_pref_chosen: nvidia/When2Call pref/DPO split (CC-BY-4.0, 9k) — the chosen honest-abstention response as a positive example. Slice… See the full description on the dataset page: https://huggingface.co/datasets/alucent/surrogate-2-honesty-sft.8b_honesty_sft_10k_v3_correct