Dataset · xtools-at
Synthetic chains of thought via trace inversion
xtools-at/Claude-Haiku-5.5-High-Reasoning-TraceInversion-3000x
Up to 10k complex English tasks with step-by-step reasoning generated using trace inversion.
What's inside
The dataset contains 1 to 10k English text examples for deep reasoning skill distillation. Tasks across 10 complex disciplines (higher mathematics, theoretical computer science, formal logic, decision theory, and scientific deduction) were generated by Gemini 3.6 Flash, and the reasoning traces were obtained using Claude Haiku 5.5 in high reasoning mode via trace inversion.
What you can do with it
- SFT fine-tuning of compact language models on complex step-by-step reasoning
- Training LoRA adapters for logic, mathematics, and computer science tasks
- Researching reasoning model distillation via trace inversion
- Tasks
- Text generation
- Languages
- English
- Size
- 1K–10K rows
- Downloads this month
- 956
- Likes
- 7
Models trained on it
- bartowski/xtools-at_Qwensploit-V2-27B-CyberSecurity-Qwen3.8-GGUFImage text → text 23K
- bartowski/xtools-at_Qwensploit-V2-9B-CyberSecurity-Qwen3.5-GGUFImage text → text 9.9K
- xtools-at/Qwensploit-V2-9B-CyberSecurity-Qwen3.5-GGUFImage text → text 4.8K
- mradermacher/Qwen3.5-4B-DeepThought-STEM-i1-GGUF 4.8K
- mradermacher/Qwen3.5-9B-DeepThought-STEM-i1-GGUF 3.6K
- xtools-at/Qwensploit-V2-27B-CyberSecurity-Qwen3.8-GGUFImage text → text 3.3K
- mradermacher/Qwen3.5-9B-DeepThought-STEM-GGUF 2.9K
- mradermacher/Qwensploit-V2-4B-CyberSecurity-Qwen3.5-i1-GGUF 2.6K
How to load it
from datasets import load_dataset
ds = load_dataset("xtools-at/Claude-Haiku-5.5-High-Reasoning-TraceInversion-3000x")