7,823 single-turn reasoning conversations from Claude Opus 4.7 reformatted for supervised fine-tuning with trl.SFTTrainer + train_on_responses_only. Each row is a single text field containing a full Qwen-style chat-template conversation.
Every conversation's assistant response (including the
... block) is output from claude-opus-4-7 with Anthropic's extended-thinking enabled. This is the… See the full description on the dataset page:
https://huggingface.co/datasets/Snow257/reasoning-distill-opus-4-7-max-sft.
