Skip to content

Dataset · Jackrong

Reconstructed Claude Opus CoT examples

Jackrong/Claude-opus-4.7-TraceInversion-5000x

5k reconstructed chain-of-thought examples from Claude 4.7-Max for SFT and DPO across multiple languages.

What's inside

The dataset contains 5,000 examples of reconstructed chain-of-thought (CoT) reasoning in the "deep logical Opus style", obtained via Trace Inversion from the Claude 4.7-Max model. The data is presented in English and several other languages (zh, ko, ru, ja, es). Readiness for SFT and DPO is stated.

What you can do with it

  • Fine-tuning LLMs on CoT via SFT
  • Using for DPO reasoning tuning
  • Evaluating model multilingual reasoning
Tasks
Text generation
Languages
English, Chinese, Korean, Russian, Japanese, Spanish
Size
1K–10K rows
Downloads this month
670
Likes
87

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("Jackrong/Claude-opus-4.7-TraceInversion-5000x")