chanwit/gemma4-opus-reasoning-10k

--- license: apache-2.0 language: - en tags: - gemma-4 - reasoning - thinking - fine-tuning - distillation size_categories: - 9K<n<10K source_datasets: - Roman1111111/claude-opus-4.6-10000x --- # gemma4-opus-reasoning-10k Gemma-4 formatted reasoning distillation dataset. ## Format This dataset uses the **Gemma-4 thinking format** with channel-based reasoning: ``` <|turn>system <|think|>You are a helpful assistant...<turn|> <|turn>user {problem}<turn|> <|turn>model <|channel>thought {step-by-step reasoning} <channel|> {final answer}<turn|> ``` ## Source Converted from Claude Opus/Qwen reasoning traces: - `Roman1111111/claude-opus-4.6-10000x` ## Statistics - **Total examples**: 9,631 - **Format**: Gemma-4 text-only - **Purpose**: Fine-tuning for step-by-step reasoning ## Usage ```python from datasets import load_dataset dataset = load_dataset("chanwit/gemma4-opus-reasoning-10k") ``` ## License Apache 2.0 (same as source datasets)许可证：Apache-2.0 语言： - 英语 标签： - gemma-4（gemma-4） - 推理 - 思考 - 微调 - 知识蒸馏 规模类别： - 9000 < 样本数 < 10000 源数据集： - Roman1111111/claude-opus-4.6-10000x # gemma4-opus-reasoning-10k 适配Gemma-4格式的推理蒸馏数据集。 ## 格式 本数据集采用**Gemma-4思考格式**，并基于通道式推理构建： <|turn>system <|think|>You are a helpful assistant...<turn|> <|turn>user {problem}<turn|> <|turn>model <|channel>thought {step-by-step reasoning} <channel|> {final answer}<turn|> ## 数据源 本数据集由Claude Opus与Qwen的推理轨迹转换而来： - `Roman1111111/claude-opus-4.6-10000x` ## 统计信息 - **总样本数**：9631 - **数据格式**：仅支持Gemma-4文本格式 - **用途**：用于分步推理能力的微调训练 ## 使用方法 python from datasets import load_dataset dataset = load_dataset("chanwit/gemma4-opus-reasoning-10k") ## 许可证 Apache 2.0（与源数据集保持一致）