CoolFace
Datasetpublic

TeichAI/lordx64-claude-opus-4.7-max-cleaned

reasoning-distill-claude-opus-4-7-max-cleaned Cleaned version of lordx64/reasoning-distill-claude-opus-4-7-max. See the original dataset for full provenance, collection methodology, and terms of use. Cleaning steps Step Filter Reason Rows removed 1 Simulated thinking (...) Rows with ... in thinking/response indicate the model learned to simulate reasoning (e.g., "Now I'm laying out the puzzle grids...") rather than actually performing it. This causes… See the full description on the dataset page: https://huggingface.co/datasets/TeichAI/lordx64-claude-opus-4.7-max-cleaned.

sourceHugging Faceapache-2.0updated 5mo agoView on Hugging Face
24likes406downloads
Dataset Card

reasoning-distill-claude-opus-4-7-max-cleaned

Cleaned version of `lordx64/reasoning-distill-claude-opus-4-7-max`.

See the original dataset for full provenance, collection methodology, and terms of use.

Cleaning steps

StepFilterReasonRows removed
1Simulated thinking (...)Rows with ... in thinking/response indicate the model learned to simulate reasoning (e.g., "Now I'm laying out the puzzle grids...") rather than actually performing it. This causes failures in agentic tasks and hallucinations during thinking.1,230 (15.1%)
2Duplicate promptsDeduplicated by exact prompt match, keeping the first occurrence.989 (12.2%)
3Missing fieldsRows without valid thinking, prompt, and response content (empty or null values).1,098 (13.5%)
Total3,317 (40.8%)
MetricValue
Original rows8,124
Final rows4,807
Retention rate59.2%

Format

Each row is a JSON object with the messages column first, following the standard chat format:

json
{
  "messages": [
    {"role": "system", "content": "You are a helpful assistant", "thinking": null},
    {"role": "user", "content": "...", "thinking": null},
    {"role": "assistant", "content": "...", "thinking": "..."}
  ],
  "system": "You are a helpful assistant",
  "prompt": "...",
  "thinking": "...",
  "response": "...",
  "model": "claude-opus-4-7"
}

License

Apache-2.0 (dataset packaging). Content subject to upstream Anthropic usage policies.