Primary sourcearXiv cs.AI
View sourceEvent ·Single source
RetroThinker框架提升语音大语言模型推理能力
RetroThinker是一个多阶段后训练框架，为Moshi模型提供自我验证和向前纠正的链式思考步骤。它通过在推理过程中结合监督微调和基于长度的直接偏好优化，在GSM8K基准测试上实现了11%的绝对准确率提升，同时保持与基线相当的延迟。
Current read
这个框架很实用，能帮语音大模型在说话时自我修正，提升推理准确率，和普通模型比效果更好。
2 sources44° AI heatLast updated 2026/9/10 17:41:53
Evidence
2 sourcesRelated source 1小互
View sourceConflict check
Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.