Google has released the Gemma 4 family of open models – from compact versions for smartphones to powerful solutions for PCs capable of competing with systems many times larger.
Google has released the Gemma 4 family of models – from compact versions for mobile devices to powerful systems capable of competing with solutions twice their size.
AI: Events
How PyTorch Achieved Faster Normalization and Its Impact on Neural Networks
Technical context • Infrastructure
The PyTorch team optimized normalization performance using the torch.compile compiler, significantly enhancing neural network training efficiency.
AI: Events
Monarch: How PyTorch Is Simplifying Supercomputer Management
Technical context • Infrastructure
PyTorch has introduced Monarch, a new tool designed to simplify the launching and debugging of distributed training tasks on large GPU clusters.
The Safetensors format has been officially adopted by the PyTorch Foundation – changing the approach to distributing model weights and securing agentic systems.
Google has introduced the fourth generation of its open-source Gemma models, ranging from compact versions for smartphones to powerful solutions for demanding computational tasks.
Researchers from MIPT have developed an AI-based system that helps design elevators for real-world use cases in specific buildings.
AI: Events
Higress: An AI Traffic Gateway to Replace the Legacy Nginx Ingress
Technical context • Infrastructure
Alibaba Cloud's Higress project has joined the CNCF Sandbox as a replacement for the retiring Nginx Ingress, offering zero downtime, support for AI traffic, and MCP servers.
Google has released Gemma 4 – a family of four open models that run directly on-device, from smartphones to laptops, without relying on the cloud.