TopologicalQwen:拓扑知识蒸馏的1.7B模型
原标题:reaperdoesntknow/TopologicalQwen
AI 摘要
Convergent Intelligence LLC 研究部门发布 TopologicalQwen,一个从 Qwen3-30B-A3B-Thinking-2507 蒸馏得到的 1.7B 参数模型。该方法称为拓扑知识蒸馏(TKD),将教师输出分布视为有界变差函数,分解为平滑、跳跃和漂移三个通道进行知识迁移,并采用 DualMind 格式训练模型进行推理、自我批判和综合回答。模型基于物理思维链数据集训练,权重以 Apache-2.0 许可开源。
正文节选
--- license: apache-2.0 library_name: transformers pipeline_tag: text-generation tags: - qwen3 - sft - trl - topological-knowledge-distillation - disc - convergent-intelligence - convergentintel - edge - distillation - knowledge-distillation base_model: - reaperdoesntknow/Qwen3-1.7B-Distilled-30B-A3B-SFT --- # TopologicalQwen **Topology-Aware Knowledge Distillation from Qwen3-30B-A3B → 1.7B** *Convergent Intelligence LLC: Research Division* --- ## What This Is TopologicalQwen is a 1.7B par