T
traeai
Sign in

公司

BAIR Blog

别名:bair

伯克利人工智能研究实验室博客

已跟踪 4 条高相关材料

TraeAI 观察

相关材料

已收录 4 条与 BAIR Blog 相关的内容,按评分排序。

Teaching LLMs to Update Beliefs for Efficient Long-Horizon Interaction

ABBEL框架通过信念状态替代完整交互历史,使LLM在长任务中保持性能,解决上下文扩展瓶颈问题。

入选理由:ABBEL使用自然语言信念状态替代完整历史记录,减少上下文负担

FeaturedArticle#LLM#长任务交互#信念状态#框架#BAIR Blog英文
From CUDA to MLX: How K-Search Brings Decades of Kernel Expertise to Apple Silicon

K-Search框架实现CUDA到MLX的自动转换,使Apple Silicon性能接近专家优化水平,Attention和Mamba SSM内核分别提升0.97x和20x。

入选理由:K-Search框架实现CUDA到MLX的自动转换,无需人工重写内核

FeaturedArticle#CUDA#MLX#Apple Silicon#K-Search#AI优化英文
Identifying Interactions at Scale for LLMs

Identifying Interactions at Scale for LLMs

BAIR Blog1512 字 (约 7 分钟)
85

Introduces SPEX and ProxySPEX algorithms to identify key interactions in large-scale LLMs.

入选理由:SPEX利用稀疏性和低度性减少计算开销

FeaturedArticle#LLM#Machine Learning#Model Interpretability中文
Gradient-based Planning for World Models at Longer Horizons

Gradient-based Planning for World Models at Longer Horizons

BAIR Blog3066 字 (约 13 分钟)
85

The article introduces GRASP, a gradient-based long-horizon planning method for world models, which improves planning robustness through virtual states, stochasticity injection, and gradient reshaping.

入选理由:GRASP通过虚拟状态实现并行优化

FeaturedArticle#Machine Learning#Reinforcement Learning#Planning Algorithm中文

跨材料问答 · BAIR Blog

回答基于:BAIR Blog 相关 4 条材料
    0 / 500

    AI may generate inaccurate information. Please verify important content.