hero
Datasets
All datasets matching “hero”prompt-perfect
Scoring popular datasets with "Self-Alignment with Instruction Backtranslation" prompt
35 datasets scored (>6B tokens)
Scoring Models used
gpt-3.5-turbo-16k
gpt-3.5-turbo-1106
gpt-3.5-turbo-0125
All datasets have 2 additional columns
score - Response from the model including CoT (if provided)
extracted_score - Extracted score from the score column as int
Datasets Scored by Prompt (Needs to be updated)… See the full description on the dataset page: https://huggingface.co/datasets/0-hero/prompt-perfect.SWE-Hero-openhands-trajectories
SWE-Hero Trajectories: Execution-based Fine-tuning for Software Engineering Agents
Data Overview
SWE-Hero Trajectories is an agentic instruction tuning dataset designed to advance the capabilities of LLMs in software engineering. This dataset comprises 34k agent
trajectories collected using the OpenHands framework. The trajectories
were synthesized using Qwen3-Coder-480B-A35B-Instruct, specifically curated for supervised fine-tuning (SFT),
aiming to improve model… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/SWE-Hero-openhands-trajectories.heronstegoattack-advbench50
StegoAttack AdvBench-50
Steganographic jailbreak data generated using the StegoAttack pipeline from the paper "Hiding in Plain Sight: A Steganographic Approach to Stealthy LLM Jailbreaks" (Geng et al., 2025).
For experiment results and analysis, see experiment.md.
What is StegoAttack?
StegoAttack is a jailbreak method that uses steganography to hide harmful queries inside benign-looking text. It embeds each word of a harmful query at a fixed position (e.g. the 2nd… See the full description on the dataset page: https://huggingface.co/datasets/heron-ai-security/stegoattack-advbench50.swe-mt-combined-coderforge-hero-lego-nex-swezero
fan-shu/swe-mt-combined-coderforge-hero-lego-nex-swezero
Concatenated mid-train dataset for Qwen3 Thinking SFT. Each source subset is loaded
in order and concatenated into a single config so one training epoch visits every
trajectory exactly once (no interleave / no oversampling).
Built from fan-shu/swe-instruct-trajectories-empty-think-inserted.
Source subsets (7)
togethercomputer__CoderForge-Preview
nvidia__SWE-Zero-openhands-trajectories
nex-agi__agent-sft… See the full description on the dataset page: https://huggingface.co/datasets/fan-shu/swe-mt-combined-coderforge-hero-lego-nex-swezero.hero_run_4_math_code