datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
proof-pileA dataset of high quality mathematical text.sob-ft-finetune-ready
SOB-FT Finetune Ready
~100k source rows → ~152k chat SFT examples for fine-tuning a small language model on JSON extraction (generation) and JSON error detection / repair (correction), with prompts aligned to our SOB extraction and zero-shot repair benchmarks.
Derived from mariem123kfg/sob-ft-extract (multi-source structured extraction corpus, excluding original SOB benchmark rows). Errors were injected in-house, then rows were materialized into ready-to-train prompt/target… See the full description on the dataset page: https://huggingface.co/datasets/seneca-center/sob-ft-finetune-ready.khayyam-challengereq2test
Req2Test
Req2Test is a large-scale, multi-language dataset for requirement-driven
unit test generation, built to enable knowledge distillation from large
teacher LLMs to small (2B–4B) student models. Depending on architecture,
precision, quantization, and runtime, such models may fit within approximately
8 GB of VRAM.
Each record pairs a natural-language functional requirement with eight
aligned code artifacts — an interface declaration and a unit test class in
each of four… See the full description on the dataset page: https://huggingface.co/datasets/centertest/req2test.synthetic_call_center_summaries
Synthetic Call Center Summaries Dataset
Overview
This dataset contains synthetic summaries of call center conversations generated by different prompt configurations.
Each record (in JSON Lines format) includes:
The original dialogue metadata.
A generated summary tailored to provide quick insights for call center service agents.
Extensive evaluation metrics and attributes such as conciseness, formatting, contextual relevance, tone, and actionability.
Intended… See the full description on the dataset page: https://huggingface.co/datasets/marccgrau/synthetic_call_center_summaries.hcms
HCMS Safety Challenge Datasets
Project repository: github.com/center-for-humans-and-machines/hcms
This repository contains two types of data: challenge conversations from an online safety challenge and synthetic conversations generated for safety evaluation.
Challenge Conversations (Online Experiment)
Raw export of 9,492 messages from 280 participants (2 duplicate ids) in the hpms_safety_challenge_main_2026_03_27 experiment. Participants were given hypothetical… See the full description on the dataset page: https://huggingface.co/datasets/center-for-humans-and-machines/hcms.