datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
qasperA dataset containing 1585 papers with 5049 information-seeking questions asked by regular readers of NLP papers, and answered by a separate set of NLP practitioners.qasper-yesnoqasper-retrieval
License & Attribution
MTEB-format derivative of allenai/qasper (QA over NLP research papers; test split). Query = question; corpus = full paper text; gold = the source paper. Licensed under CC-BY-4.0 (same as source).
qasper-rewrite-gpt-4.1task461_qasper_question_generation
Dataset Card for Natural Instructions (https://github.com/allenai/natural-instructions) Task: task461_qasper_question_generation
Additional Information
Citation Information
The following paper introduces the corpus in detail. If you use the corpus in published work, please cite it:
@misc{wang2022supernaturalinstructionsgeneralizationdeclarativeinstructions,
title={Super-NaturalInstructions: Generalization via Declarative Instructions on 1600+ NLP Tasks}… See the full description on the dataset page: https://huggingface.co/datasets/Lots-of-LoRAs/task461_qasper_question_generation.qasper-rag
QASPER RAG
Dataset for Retrieval-Augmented Generation (RAG) based on QASPER.
Structure
Subset
Splits
Description
corpus
train (default)
Paper chunks (abstract + full-text paragraphs) shared across all query splits
queries
train, dev, test
Information-seeking questions over scientific papers
qrels
train, dev, test
Relevance judgments (query ↔ paragraph chunk)
answers
train, dev, test
Reference answers (longest valid free-form answer)
top_ranked… See the full description on the dataset page: https://huggingface.co/datasets/DinoStackAI/qasper-rag.DCAgent2_terminal_bench_2_mlfoundations-dev_qasper-sandboxes-traces-terminus-2_2cddc52dQASPERChunkRetrievalqasperpaper-qa-qasper-sfttask460_qasper_answer_generation
Dataset Card for Natural Instructions (https://github.com/allenai/natural-instructions) Task: task460_qasper_answer_generation
Additional Information
Citation Information
The following paper introduces the corpus in detail. If you use the corpus in published work, please cite it:
@misc{wang2022supernaturalinstructionsgeneralizationdeclarativeinstructions,
title={Super-NaturalInstructions: Generalization via Declarative Instructions on 1600+ NLP Tasks}… See the full description on the dataset page: https://huggingface.co/datasets/Lots-of-LoRAs/task460_qasper_answer_generation.dcagent2-terminal-bench-2-mlfoundations-dev-qasper-sandboxes-traces-terminus-2-16962780QASPER
QASPER
An MTEB dataset
Massive Text Embedding Benchmark
LMEB semantic retrieval task based on QASPER, retrieving evidence passages from research papers for information-seeking questions.
Task category
Retrieval (text-to-text)
Domains
Academic, Written
Reference
LMEB: Long-horizon Memory Embedding Benchmark
Source datasets:
KaLM-Embedding/LMEB
How to evaluate on this task
You can evaluate an embedding model on this dataset using the following… See the full description on the dataset page: https://huggingface.co/datasets/mteb/QASPER.qasper
Preprocessed QASPER dataset
Working doc: https://docs.google.com/document/d/1gYPhPNJ5LGttgjix1dwai8pdNcqS6PbqhsM7W0rhKNQ/edit?usp=sharing
Original:
Dataset: https://github.com/allenai/qasper-led-baseline
Baseline repo: https://github.com/allenai/qasper-led-baseline
HF: https://huggingface.co/datasets/allenai/qasper
Differences of our implementation over the original implementation:
We use the dataset provided at https://huggingface.co/datasets/allenai/qaspersince it doesn't… See the full description on the dataset page: https://huggingface.co/datasets/albertgong1/qasper.delta-mem-qasper-data
Introduction
This repository contains the δ-mem training data, as presented in the paper δ-mem: Efficient Online Memory for Large Language Models.
δ-mem is a lightweight online memory mechanism that augments a frozen backbone with a compact associative memory state. It projects context into a low-dimensional space and updates a state matrix via delta-rule learning, allowing for efficient long-term memory utilization without full fine-tuning or context extension.
Paper:… See the full description on the dataset page: https://huggingface.co/datasets/huaXiaKyrie/delta-mem-qasper-data.converted_qasper
Dataset Card for "converted_qasper"
More Information needed
qasper
Dataset Card for "qasper"
More Information needed
qasper-squad
Quasper into squad version
This is a change of format of qasper dataset into squad format.
qasper-sandboxes_glm_4.7_traces_jupiterqwen-qasper-subkv-sweep
QASPER end-to-end sub-KV cache sweep
This dataset contains the complete outputs of the examples/e2e_subkv_sweep
experiment from gated-continual-cartridges, run on 2026-09-18 at git commit
72e91560eac9ceb22eba495dead4710fe25cc0f1.
The sweep uses Qwen3-4B-Instruct-2507 and QASPER, with five sequential phases
(QA, MT, SA, ASR, KG). Each arm uses top_t / cache_budget = 6.25%,
ridge_lambda = 0, and delta_lambda = 0.01. The reported metric is
teacher-forced mean token loss in… See the full description on the dataset page: https://huggingface.co/datasets/qtris123/qwen-qasper-subkv-sweep.DCAgent2_terminal_bench_2_mlfoundations-dev_qasper-sandboxes-traces-terminus-2_7a63da23dcagent2-terminal-bench-2-mlfoundations-dev-qasper-sandboxes-traces-terminus-2-92363655SARA-QASPER
SARA QASPER (reformatted)
Reformatted QASPER data used by
SARA: Selective and Adaptive Retrieval-augmented Generation with Context Compression (ACL 2026,
arXiv:2507.05633). Code: Ahren09/SARA.
The SARA Quick Start (python -m src.data.make_qasper_splits) downloads this dataset automatically;
you can also load it directly:
from datasets import load_dataset
qa = load_dataset("Ahren09/SARA-QASPER", "qa") # train / test
align =… See the full description on the dataset page: https://huggingface.co/datasets/Ahren09/SARA-QASPER.converted_qasper
Dataset Card for "converted_qasper"
More Information needed
qasper_t5DCAgent_dev_set_71_tasks_mlfoundations-dev_qasper-sandboxes-traces-terminus-2_203fab8d58DCAgent2_terminal_bench_2_mlfoundations-dev_qasper-sandboxes-traces-terminus-2_ea0a70edenvoy-qasper-code-trajectories
Envoy QASPER Code-Execution Trajectory Pilot
This is a small, fully disclosed pilot of executable research-agent trajectories.
Claude Sonnet 5 generated Python actions against a persistent document REPL. The
Envoy pipeline executed every action and retained the real observations. An AI
coding assistant then reviewed answer support, stopping behavior, and replay.
This release is useful for studying trajectory validation and citation failures.
It is not a production-ready SFT… See the full description on the dataset page: https://huggingface.co/datasets/jasonlingg/envoy-qasper-code-trajectories.terminal_bench_2_a1_qasper_20260711_155915qasperA dataset containing 1585 papers with 5049 information-seeking questions asked by regular readers of NLP papers, and answered by a separate set of NLP practitioners.