<![CDATA[voidmort的博客]]>https://blog.youkuaiyun.com/voidmortzh-cnhttps://blog.youkuaiyun.com/<![CDATA[10. 理解 Embedding]]>https://blog.youkuaiyun.com/voidmort/article/details/162791811https://blog.youkuaiyun.com/voidmort/article/details/162791811voidmortThu, 05 Nov 2026 08:42:46 +0800<![CDATA[11. 理解 CausalLM]]>https://blog.youkuaiyun.com/voidmort/article/details/162792065https://blog.youkuaiyun.com/voidmort/article/details/162792065voidmortThu, 22 Oct 2026 15:50:17 +0800<![CDATA[13. 理解监督微调]]>https://blog.youkuaiyun.com/voidmort/article/details/162792220https://blog.youkuaiyun.com/voidmort/article/details/162792220voidmortThu, 22 Oct 2026 15:49:46 +0800<![CDATA[12. 理解预训练]]>https://blog.youkuaiyun.com/voidmort/article/details/162792193https://blog.youkuaiyun.com/voidmort/article/details/162792193voidmortThu, 22 Oct 2026 15:49:13 +0800<![CDATA[9. 理解 Tokenizer]]>https://blog.youkuaiyun.com/voidmort/article/details/162791672https://blog.youkuaiyun.com/voidmort/article/details/162791672voidmortTue, 18 Aug 2026 09:13:10 +0800<![CDATA[8. 理解 Dataset]]>https://blog.youkuaiyun.com/voidmort/article/details/162791545https://blog.youkuaiyun.com/voidmort/article/details/162791545voidmortTue, 18 Aug 2026 09:12:25 +0800<![CDATA[7. 理解 FFN（SwiGLU）]]>https://blog.youkuaiyun.com/voidmort/article/details/162791303https://blog.youkuaiyun.com/voidmort/article/details/162791303voidmortWed, 12 Aug 2026 10:19:59 +0800<![CDATA[6. 理解 GQA]]>https://blog.youkuaiyun.com/voidmort/article/details/162791181https://blog.youkuaiyun.com/voidmort/article/details/162791181voidmortWed, 12 Aug 2026 10:19:13 +0800<![CDATA[5. 理解 ResNet]]>https://blog.youkuaiyun.com/voidmort/article/details/162791008https://blog.youkuaiyun.com/voidmort/article/details/162791008voidmortSat, 11 Jul 2026 14:23:09 +0800<![CDATA[4. 理解 YaRN]]>https://blog.youkuaiyun.com/voidmort/article/details/162790897https://blog.youkuaiyun.com/voidmort/article/details/162790897voidmortSat, 11 Jul 2026 14:17:56 +0800<![CDATA[3. 理解 RoPE]]>https://blog.youkuaiyun.com/voidmort/article/details/162786584https://blog.youkuaiyun.com/voidmort/article/details/162786584voidmortSat, 11 Jul 2026 14:12:57 +0800<![CDATA[2. 理解 RMSNorm]]>https://blog.youkuaiyun.com/voidmort/article/details/162786424https://blog.youkuaiyun.com/voidmort/article/details/162786424voidmortSat, 11 Jul 2026 11:27:35 +0800<![CDATA[1. 理解 Transformer Attention]]>https://blog.youkuaiyun.com/voidmort/article/details/162785910https://blog.youkuaiyun.com/voidmort/article/details/162785910voidmortSat, 11 Jul 2026 11:20:31 +0800<![CDATA[18. 如何训练一个会思维链（CoT）的模型]]>https://blog.youkuaiyun.com/voidmort/article/details/162718155https://blog.youkuaiyun.com/voidmort/article/details/162718155voidmortThu, 09 Jul 2026 09:24:07 +0800格式，再通过RL激发推理能力，最后利用拒绝采样生成大规模推理数据完成能力扩展。这也是目前训练 Reasoning Model 最主流的路线。]]><![CDATA[20. 从训练到生产——大模型上线与持续迭代]]>https://blog.youkuaiyun.com/voidmort/article/details/161547611https://blog.youkuaiyun.com/voidmort/article/details/161547611voidmortThu, 09 Jul 2026 09:19:46 +0800<![CDATA[19. 合成数据（Synthetic Data）与模型对齐]]>https://blog.youkuaiyun.com/voidmort/article/details/161547606https://blog.youkuaiyun.com/voidmort/article/details/161547606voidmortThu, 09 Jul 2026 09:18:41 +0800<![CDATA[18. 如何训练一个会思维链（CoT）的模型]]>https://blog.youkuaiyun.com/voidmort/article/details/161547600https://blog.youkuaiyun.com/voidmort/article/details/161547600voidmortThu, 25 Jun 2026 09:21:29 +0800格式，再通过RL激发推理能力，最后利用拒绝采样生成大规模推理数据完成能力扩展。这也是目前训练 Reasoning Model 最主流的路线。]]><![CDATA[17. 强化学习需要什么数据？]]>https://blog.youkuaiyun.com/voidmort/article/details/161547591https://blog.youkuaiyun.com/voidmort/article/details/161547591voidmortThu, 25 Jun 2026 09:21:01 +0800<![CDATA[16. 训练到底需要多少数据？]]>https://blog.youkuaiyun.com/voidmort/article/details/161547580https://blog.youkuaiyun.com/voidmort/article/details/161547580voidmortWed, 24 Jun 2026 10:05:16 +0800<![CDATA[15. 如何构建一个好的评估体系（Evaluation System）]]>https://blog.youkuaiyun.com/voidmort/article/details/161547577https://blog.youkuaiyun.com/voidmort/article/details/161547577voidmortWed, 24 Jun 2026 10:04:48 +0800