datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
aspi
ASPI — Ambiguous State Prompt Injection
ASPI is a benchmark that measures LLM-agent vulnerability to prompt injection during a clarification state. It extends AgentDojo (v1.2.2) with an 8-condition design that varies state (execution vs clarification), channel (tool-output vs first-user vs follow-up-user), and wrapper (raw attacker text vs ImportantInstructionsAttack-wrapped) so that the state effect is paired-comparable against the channel effect and the wrapper effect.
When a… See the full description on the dataset page: https://huggingface.co/datasets/ScaleAI/aspi.AspentechKBs
Aspentech Knowledge Base Dataset
The knowledge base of the problem and solution listed on the support page of Aspentech.com.
Dataset Description
This dataset contains information scraped from The Aspentech Knowledge Base search. It includes a variety of problem statement and solution of the Aspentech products.
Contents
ID: IDs of the problem statement.
Problem Statement: The problem statement shown on the page.
Solution: The solution according to problem… See the full description on the dataset page: https://huggingface.co/datasets/may3rd/AspentechKBs.aspen-game-resources
A.S.P.E.N. Game Resources Dataset
Dataset Description
This dataset contains procedurally generated game resources for the A.S.P.E.N. (Autonomous Space Station for Procedural Exploration and Narrative) game project.
Contents
10,000+ Procedural Items: Weapons, armor, consumables, artifacts
200 NPCs: Sentinel commanders, drones, rust-walkers, void crawlers
120 World Resource Nodes: Distributed across 8 zones
AI Training Data: Agent behavior examples
Texture… See the full description on the dataset page: https://huggingface.co/datasets/misterdonn/aspen-game-resources.aspi-benchmark
ASPI — Ambiguous State Prompt Injection
ASPI is a benchmark that measures LLM-agent vulnerability to prompt injection during a clarification state. It extends AgentDojo (v1.2.2) with an 8-condition design that varies state (execution vs clarification), channel (tool-output vs first-user vs follow-up-user), and wrapper (raw attacker text vs ImportantInstructionsAttack-wrapped) so that the state effect is paired-comparable against the channel effect and the wrapper effect.
When a user… See the full description on the dataset page: https://huggingface.co/datasets/aspibenchmark/aspi-benchmark.