PyTorch BlogThe EUROPA Union successfully trained a MoE model with 951.9B parameters on the JUPITER supercomputer using PyTorch and TorchTitan, increasing single GPU throughput from 90 to approximately 1,300 tokens/s with 384 NVIDIA GH200 GPUs.
vLLM BlogFeatured7272NVIDIA Research and Inferact's REINFORCE Zero proposes to first align rollout with training kernel, then use unclipped token-level importance sampling to correct weight lag, achieving a training step throughput 2.40 times that of synchronous training under 16 GPUs and 12K updates.
Why it matters: The paper breaks down the gap between asynchronous RL's rollout and training into two items: weight lag and kernel inconsistency, and proposes an alignment-free, no-trimming importance sampling method after alignment.
NVIDIA BlogC-BRAIN and NVIDIA are using AI to change the pace of dementia research. They are accelerating the development of new therapies by integrating previously unpublished Alzheimer's research data, including unpublished results and controlled proprietary data. C-BRAIN employs a hybrid centralized and federated architecture to protect data privacy, deploying 64 NVIDIA data center GPUs and cloud nodes, and plans to introduce the NVIDIA Blackwell system to support its expansion goals.
Prime Intellect BlogPrime Intellect releases the SYNTHETIC-2 dataset, containing four million verified reasoning trajectories, covering the most comprehensive complex reinforcement learning tasks and validators to date. The dataset is generated through a pipeline parallel distributed reasoning collaboration involving global computing contributors, with 1,253 GPUs (including 49 8xH200 nodes, 43 8xH100 nodes, and a large number of consumer-grade 3090 and 4090 GPUs) participating in its creation over 3 days.
Prime Intellect BlogDeepSeek releases SYNTHETIC-2, which includes 20 complex reinforcement learning tasks and verification reasoning trajectories, supporting heterogeneous computing from consumer-grade GPUs to super-large-scale clusters, using the DeepSeek-R1-0528 model to generate data and label difficulty.
Prime Intellect Blog3838Prime Mega MoE enables the DeepSeek Mega MoE kernel to be used for training by integrating backpropagation kernels, achieving up to 3.11 times faster MoE kernel speed compared to the default PyTorch implementation, and 1.34 times faster end-to-end training speed compared to the Prime-RL baseline.
Import AI4242Humanity’s Sixth Sense review shows GPT 6 Astra scores 53.6 in intuitive visual reasoning tasks with a pass@1 rate, far below the human baseline of 93.1, with the open-source Chinese model GLM-5.3-flash scoring 26.8 and Kimi K3 scoring 25.5.
Microsoft ResearchGigaPath-Flash and GigaTIME-Flash significantly reduce computational requirements through streamlined ViT-S encoders, making large-scale pathological research more feasible while maintaining performance.
