VFlash treats visual planning as a fast, System-1 capability, pointing to a shift in how embodied agents handle low-latency decision-making instead of reasoning purely through text. Robotics and physical AI developers can review the project page and the lead's thread to understand the architecture details. It comes from the same NVIDIA research group as Jim Fan's GEAR Lab, whose embodied agent work is a useful benchmark for comparison.

#VFlash #NVIDIA #Physical AI #multimodal

Event

Jim Fan@DrJimFan8 sourcesSource

Industry BuildersMulti-source01:07

Chinese contentGergely Orosz 对比 NVIDIA 与 Apple 员工 AI 使用政策

Orosz 晒了个有意思的对比：NVIDIA 员工每月 1 万美元 AI 额度随便用，Apple 却不鼓励用 AI，谁的做法更聪明可以聊聊

#NVIDIA #Apple #Gergely Orosz #AI应用

Event

Gergely Orosz@GergelyOrosz11 sourcesSource

How-to BuildersMulti-sourceSelected01:02

NVIDIA shows grouped execution for MoE optimization in PyTorch

The non-obvious point is that the bottleneck in MoE training is often kernel launch overhead from Python-level expert loops, not the expert math itself. Engineers training MoE models in PyTorch can apply grouped execution to cut that overhead without restructuring their expert weights. The worked example uses biological foundation models, a niche domain where sparse expert routing is common.

#PyTorch #NVIDIA #MoE #fine-tuning

Event

官方账号PyTorch@PyTorch7 sourcesSource

How-to BuildersMulti-source00:40

Chinese contentNVIDIA 拆解 5 种 AI 内存优化方法：从压缩模型权重到跨服务器复用上下文

NVIDIA 用一个视频讲了 5 种省显存的思路，做推理部署的可以看看，里面有跨服务器复用上下文这种偏系统的做法

#NVIDIA #推理优化 #模型权重 #KV Cache

Event

官方账号NVIDIA AI@NVIDIAAI6 sourcesSource

Products BuildersMulti-sourceSelected13:30

NVIDIA releases CUDA Toolkit 13.5 with Vera Rubin GA support

The GA of Vera Rubin tooling signals that NVIDIA's next-generation platform is now ready for enterprise deployment, so infrastructure teams can begin planning migrations well ahead of hardware ramp-up. The MPS v3 scheduling and SM-partitioning changes matter most for operators running multi-tenant GPU clusters, since they enable finer control of compute sharing without per-shape autotuning. Compared with relying on CUDA 13.x preview builds, this release offers the first stable software path for Vera Rubin training and inference workloads.

#NVIDIA #CUDA #Vera Rubin #GPU

Event

IT之家11 sourcesSource