[LG]《Pre-training under infinite compute
[LG]《Pre-training under infinite compute
爱生活爱珂珂的文章
![[LG]《Pre-training under infinite compute](http://dayu-img.uc.cn/columbus/img/oc/1002/be631c3f9829d09cd087406b1f0dd7f7.jpg)
![[LG]《Self-Improving Embodied Foundation](http://dayu-img.uc.cn/columbus/img/oc/1002/27fe42d732a1840e303a7d75b600e0e3.jpg)
[LG]《Self-Improving Embodied Foundation
[LG]《Self-Improving Embodied Foundation
![[CL]《LLM-JEPA: Large Language Models Mee](http://dayu-img.uc.cn/columbus/img/oc/1002/5bc963aa0816fb193d16ee7f082a6a07.jpg)
[CL]《LLM-JEPA: Large Language Models Mee
[CL]《LLM-JEPA: Large Language Models Mee
![早！[太阳] 早安 ](http://dayu-img.uc.cn/columbus/img/oc/1002/75980e37e3e37a8ace176f1e9ecfb454.jpg)
早！[太阳] 早安
早！[太阳] 早安
Mistral 推出全新推理模型 Magistral 1.2，性能再度跃升。•
Mistral 推出全新推理模型 Magistral 1.2，性能再度跃升。•
《GPU utilisation and performance improve
《GPU utilisation and performance improve
GPU 编程入门资源精选，助你快速突破学习瓶颈：• JAX 官方 GPU 指南详
GPU 编程入门资源精选，助你快速突破学习瓶颈：• JAX 官方 GPU 指南详
![[人人能懂] 从熔炼答案、系统权衡到逻辑自洽有没有想过，AI高手不是靠找答案，而](http://dayu-img.uc.cn/columbus/img/oc/1002/221cf6b5541c1e2baac02b2ec91e5769.jpg)
[人人能懂] 从熔炼答案、系统权衡到逻辑自洽有没有想过，AI高手不是靠找答案，而
[人人能懂] 从熔炼答案、系统权衡到逻辑自洽有没有想过，AI高手不是靠找答案，而
![[LG]《Teaching LLMs to Plan: Logical Chai](http://dayu-img.uc.cn/columbus/img/oc/1002/4df169c1edfdd082cbc962474f8539a9.jpg)
[LG]《Teaching LLMs to Plan: Logical Chai
[LG]《Teaching LLMs to Plan: Logical Chai
![[CL]《Early Stopping Chain-of-thoughts in](http://dayu-img.uc.cn/columbus/img/oc/1002/8974aff14ef4c1aa33c64b807cc404d8.jpg)
[CL]《Early Stopping Chain-of-thoughts in
[CL]《Early Stopping Chain-of-thoughts in
![[LG]《Imagined Autocurricula》A H. Güzel,](http://dayu-img.uc.cn/columbus/img/oc/1002/dedb5b4ab912cb02ac2208b95f347fc8.jpg)
[LG]《Imagined Autocurricula》A H. Güzel,
[LG]《Imagined Autocurricula》A H. Güzel,
![[LG]《SteeringControl: Holistic Evaluatio](http://dayu-img.uc.cn/columbus/img/oc/1002/865155975e3e348eb1e0ae38d6b67319.jpg)
[LG]《SteeringControl: Holistic Evaluatio
[LG]《SteeringControl: Holistic Evaluatio
![[LG]《Compute as Teacher: Turning Inferen](http://dayu-img.uc.cn/columbus/img/oc/1002/215f85ffca6eba41854894c78e181206.jpg)
[LG]《Compute as Teacher: Turning Inferen
[LG]《Compute as Teacher: Turning Inferen
![早！[太阳] 早安 ](http://dayu-img.uc.cn/columbus/img/oc/1002/5d810132faf9ede6f95d32b448e1dd16.jpg)
早！[太阳] 早安
早！[太阳] 早安
ATLAS 架构实现了超长上下文处理能力，开辟了大规模语言模型的新边界。• 创新
ATLAS 架构实现了超长上下文处理能力，开辟了大规模语言模型的新边界。• 创新
Claude Code 深度开发智能指南，开启开发效率与智能协同新纪元：• 集成
Claude Code 深度开发智能指南，开启开发效率与智能协同新纪元：• 集成
在 Groq 的 ML Engineer 面试中，衡量 LLM 推理性能绝非单一
在 Groq 的 ML Engineer 面试中，衡量 LLM 推理性能绝非单一
MemoRAG开创了基于记忆的RAG框架，突破传统只针对显式查询的限制，实现对海
MemoRAG开创了基于记忆的RAG框架，突破传统只针对显式查询的限制，实现对海
![[LG]《Reasoning Models Can be Accurately](http://dayu-img.uc.cn/columbus/img/oc/1002/3e76cf3e7ea708432b884787758ebd51.jpg)
[LG]《Reasoning Models Can be Accurately
[LG]《Reasoning Models Can be Accurately
![[LG]《RL Fine-Tuning Heals OOD Forgetting](http://dayu-img.uc.cn/columbus/img/oc/1002/0264d7a46fdb036a10eba24963212ba8.jpg)
[LG]《RL Fine-Tuning Heals OOD Forgetting
[LG]《RL Fine-Tuning Heals OOD Forgetting
作者信息
