datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Eklav-Math-CotGen-Data
HETU-MathReasoning-CotGen-Data
Training data for the HETU (Hints Enable True Understanding) paper.
Task: math reasoning (AIME, GSM8K, MATH-500, Omni-MATH, GPQA-Diamond, MMLU)
Method: CotGen
Examples: 3,481 train / 35-36 held-out val
Format: ShareGPT (system + conversations: [{from, value}]), used for LoRA SFT via LLaMA-Factory.
Single-turn ShareGPT conversations, curated DeepSeek-R1-style math reasoning distillation. Each row: the raw math/logic problem (human turn) and the… See the full description on the dataset page: https://huggingface.co/datasets/AdarshSingh7647/Eklav-Math-CotGen-Data.Eklav-Math-Data
HETU-MathReasoning-CotCond-Data
Training data for the HETU (Hints Enable True Understanding) paper.
Task: math reasoning (AIME, GSM8K, MATH-500, Omni-MATH, GPQA-Diamond, MMLU)
Method: CotCond
Examples: 3,481 train / 35-36 held-out val
Format: ShareGPT (system + conversations: [{from, value}]), used for LoRA SFT via LLaMA-Factory.
Single-turn ShareGPT conversations. Each row: the math/logic problem plus a partial excerpt of the teacher's reasoning as a hint (human turn), and… See the full description on the dataset page: https://huggingface.co/datasets/AdarshSingh7647/Eklav-Math-Data.Eklav-Reranker-Data
HETU-PassageReranking-CotCond-Data
Training data for the HETU (Hints Enable True Understanding) paper.
Task: passage reranking (BRIGHT / NevIR benchmarks)
Method: CotCond
Examples: 381,934 train / held-out val
Format: ShareGPT (system + conversations: [{from, value}]), used for LoRA SFT via LLaMA-Factory.
Single-turn ShareGPT conversations. Each row: a query+passage relevance-judgment prompt plus a partial reasoning hint (human turn), and a continuation plus true/false… See the full description on the dataset page: https://huggingface.co/datasets/AdarshSingh7647/Eklav-Reranker-Data.Eklav-Reranker-CotGen-Data
HETU-PassageReranking-CotGen-Data
Training data for the HETU (Hints Enable True Understanding) paper.
Task: passage reranking (BRIGHT / NevIR benchmarks)
Method: CotGen
Examples: 381,934 train / held-out val
Format: ShareGPT (system + conversations: [{from, value}]), used for LoRA SFT via LLaMA-Factory.
Single-turn ShareGPT conversations. Each row: a query+passage relevance-judgment prompt (human turn) and the model's full chain-of-thought plus true/false judgment (gpt turn)… See the full description on the dataset page: https://huggingface.co/datasets/AdarshSingh7647/Eklav-Reranker-CotGen-Data.Eklav-Reranker-AnswerOnly-Data
Eklav-Reranker-AnswerOnly-Data
Training data for the Eklav paper.
Task: passage reranking (BRIGHT / NevIR benchmarks)
Method: Answer-only (no reasoning of any kind -- the no-CoT floor)
Examples: 381,934 train / 3,857 held-out val
Format: ShareGPT (system + conversations: [{from, value}]), used for LoRA SFT via LLaMA-Factory.
Single-turn ShareGPT conversations. Each row: a query+passage relevance-judgment prompt (human turn) and a bare true/false judgment (gpt turn) -- no hint… See the full description on the dataset page: https://huggingface.co/datasets/AdarshSingh7647/Eklav-Reranker-AnswerOnly-Data.