datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
vsr_random
VSR: Visual Spatial Reasoning
This is the random set of VSR: Visual Spatial Reasoning (TACL 2023) [paper].
Usage
from datasets import load_dataset
data_files = {"train": "train.jsonl", "dev": "dev.jsonl", "test": "test.jsonl"}
dataset = load_dataset("cambridgeltl/vsr_random", data_files=data_files)
Note that the image files still need to be downloaded separately. See data/ for details.
Go to our github repo for more introductions.
Citation
If you find VSR… See the full description on the dataset page: https://huggingface.co/datasets/cambridgeltl/vsr_random.Java_method2test_chatml
Java Method to Test ChatML
This dataset is based on the methods2test dataset from Microsoft. It follows the ChatML template format: [{'role': '', 'content': ''}, {...}].
Originally, methods2test contains only Java methods at different levels of granularity along with their corresponding test cases. The different focal method segmentations are illustrated here:
To simulate a conversation between a Java developer and an AI assistant, I introduce two key parameters:
The prompt… See the full description on the dataset page: https://huggingface.co/datasets/random-long-int/Java_method2test_chatml.Amazon-combined
Amazon Combined Dataset
E-commerce dataset that combines metadata, reviews, and sample question/answer pairs. combined.json contains the dataset and user2asin.json contains a file that maps user_id from reviews to an ASIN for capturing user preferences.
Data Fields
Field
Type
Explanation
main_category
str
Main category (i.e., domain) of the product.
title
str
Name of the product.
average_rating
float
Rating of the product shown on the product page.… See the full description on the dataset page: https://huggingface.co/datasets/randomath/Amazon-combined.MMLU_ExpertPrompt_Random_01This is a massive multitask test consisting of multiple-choice questions from various branches of knowledge, covering 57 tasks including elementary mathematics, US history, computer science, law, and more.MutQA
CrossValQA
CrossValQA is a cross-validated, sentence-grounded question-answering
dataset about genetic mutations, constructed from full-text PubMed articles.
Every record links a natural-language question to a specific variant, a
specific PubMed article, and a specific cited sentence span, and every
answer was produced by two independent LLMs that had to agree before the
record was admitted.
The released train/test splits (homology and random configs)
contain only cross-grounded… See the full description on the dataset page: https://huggingface.co/datasets/random987654321/MutQA.GSM8K-Random-All
GSM8K-Random-All
A dataset for training LLMs with random backtracking capabilities. This dataset augments the original GSM8K math word problems with synthetic error injection and backtrack recovery sequences.
Overview
This dataset teaches models to:
Make "mistakes" (random error tokens)
Recognize the mistake
Use <|BACKTRACK|> tokens to "delete" the errors
Continue with the correct solution
Backtracking Mechanism
The <|BACKTRACK|> token functionally acts as a… See the full description on the dataset page: https://huggingface.co/datasets/mtybilly/GSM8K-Random-All.MMLU_ExpertPrompt_Random_03This is a massive multitask test consisting of multiple-choice questions from various branches of knowledge, covering 57 tasks including elementary mathematics, US history, computer science, law, and more.MMLU_ExpertPrompt_Random_02This is a massive multitask test consisting of multiple-choice questions from various branches of knowledge, covering 57 tasks including elementary mathematics, US history, computer science, law, and more.MMLU_ExpertPrompt_RandomMedicalQnA-llama2
MedicalQnA-llama2 Dataset
This repository contains the MedQuad-MedicalQnADataset specifically formatted for use with the LLaMA 2 prompt template. The dataset consists of medical questions categorized by question type, along with their corresponding answers. It is designed for text-to-text generation and text-generation tasks, particularly focusing on the medical domain.
Dataset Structure
The dataset is structured to include prompts that follow the LLaMA 2 template. Each… See the full description on the dataset page: https://huggingface.co/datasets/randomani/MedicalQnA-llama2.Random_CombinedMultilingual-Thinking-SR
Multilingual Thinking Dataset - Serbian Translation
This dataset is a Serbian translation of the HuggingFaceH4/Multilingual-Thinking dataset, created for fine-tuning GPT OSS 20b model with Serbian language capabilities.
Dataset Description
This dataset contains conversation messages with chain-of-thought reasoning, translated from English to Serbian in both Latin and Cyrillic scripts.
Translation Methodology
Source: English rows from the original… See the full description on the dataset page: https://huggingface.co/datasets/randomanon000/Multilingual-Thinking-SR.Bahamut_random_topic來自巴哈姆特論壇的隨機討論串
測試模型訓練效果
randomized-yarn-babilong
Randomized YaRN — BABILong (QA3)
Frozen train/eval splits for the BABILong QA3 experiments in Randomized YaRN Improves
Length Generalization for Long-Context Reasoning (Mehta, Yin, Durrett). Published so the
paper's numbers reproduce independently of library/tokenizer versions — load these
fixed bytes instead of regenerating.
Splits
split
rows
description
train
4000
1000 subsampled QA3 examples per bin (0k/2k/4k/8k), seed 42
test
2703
the paper's… See the full description on the dataset page: https://huggingface.co/datasets/Mannyboyy/randomized-yarn-babilong.