datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
DeepSeek-v3.1-reasoner-Distilled-math-samples
DeepSeek-V3.1 Distillation with NVIDIA Nemotron-Post-Training-Dataset-v2 (Math Subset)
The release of DeepSeek-V3.1 has attracted wide attention in the AI community. Its significant improvements in reasoning ability provide a new opportunity to explore optimization of domain-specific models. To investigate the potential of this model in complex mathematical reasoning tasks, I selected the math subset from NVIDIA’s newly released Nemotron-Post-Training-Dataset-v2 as seed problems and… See the full description on the dataset page: https://huggingface.co/datasets/Jackrong/DeepSeek-v3.1-reasoner-Distilled-math-samples.cancer-screening-evidence-reasoner
Cancer Screening Evidence Reasoner (AutoScientist Challenge)
Fine-tuning dataset for teaching a language model to answer cancer screening eligibility and evidence questions with exact, verifiable citations — not hedged guesses.
Motivation
Base models know screening guidelines roughly but invent citations and get exact statistics wrong. Every completion in this dataset is computed by a rule engine from verified USPSTF and SEER ground truth — not LLM-generated.… See the full description on the dataset page: https://huggingface.co/datasets/vnytht/cancer-screening-evidence-reasoner.nemotron-eco-reasoner-v14
Nemotron Eco Reasoner — Dataset v14
Training dataset (9,686 records) for a LoRA adapter on
NVIDIA-Nemotron-3-Nano-30B-A3B-BF16, built for the NVIDIA Nemotron Model
Reasoning Challenge.
Each record is a chat-format example: a system prompt, a user puzzle, and an
assistant response ending in </think> then \boxed{<answer>}.
What is new in v14
v14 keeps the exact per-category composition of the best prior dataset (v8,
the 0.67-Kaggle record) but replaces the… See the full description on the dataset page: https://huggingface.co/datasets/alrobles/nemotron-eco-reasoner-v14.parason-data
parason-data
Evaluation traces and structural-analysis artifacts for the parallel-reasoning line of work.
Training data is not here — it lives in parallel-reasoner/sft-ours (splits 1x, 8x).
This repo holds generated traces, so that structural claims about model behaviour can be
re-derived rather than taken on trust.
Layout
aime24/<model-name>/traces.jsonl
aime24/Qwen3-8B-sft-ours8x-ar/
Traces from parallel-reasoner/Qwen3-8B-sft-ours8x-ar
— the… See the full description on the dataset page: https://huggingface.co/datasets/parallel-reasoner/parason-data.news-summarizer-reasoner
News Summarizer with Reasoning
Overview
This dataset is designed for news summarization tasks, featuring both original news articles and their corresponding summaries. The dataset includes a 'news' column sourced from three different datasets, along with 'cleaned_summary' and 'cleaned_reasoning' columns generated using large language models.
Dataset Structure
news: Original news articles.
cleaned_summary: Summaries of the news articles in bullet points.… See the full description on the dataset page: https://huggingface.co/datasets/seacorn/news-summarizer-reasoner.