PyTorch Conference

llm-d 1.0: Open Source LLM Inference Scaling Hits Production Grade

llm-d 1.0: Open Source LLM Inference Scaling Hits Production Grade

AGNTCon + MCPCon North America 2026, AI, Application Development, DevOps, Events, Open Source, PyTorch Conference, Technology /

The llm-d project has reached 1.0, bringing production-grade LLM inference scaling to Kubernetes-native infrastructure under CNCF governance. With a blended serving cost of $0.11 per million tokens and support for 3,000 concurrent coding agents, the release changes the economics of enterprise AI deployment. ECI Research data shows why developer velocity and AI adoption friction make this the right infrastructure bet at the right moment.

llm-d 1.0: Open Source LLM Inference Scaling Hits Production Grade Read More »

Google TorchTPU: Native PyTorch for Cloud TPUs Goes Open Source

Google TorchTPU: Native PyTorch for Cloud TPUs Goes Open Source

AGNTCon + MCPCon North America 2026, AI, Application Development, Cloud, DevOps, Events, Infrastructure, Machine Learning, Open Source, PyTorch Conference /

Google announced TorchTPU at PyTorch Conference 2026, open sourcing native PyTorch support for Cloud TPUs and integrating it across vLLM, SGLang, llm-d, Ray, and Hugging Face. The release is less a hardware story than a developer ecosystem play, designed to make TPUs the path of least resistance for AI teams already building on PyTorch. LiteRT also joins the PyTorch Ecosystem, closing the training-to-edge deployment gap.

Google TorchTPU: Native PyTorch for Cloud TPUs Goes Open Source Read More »

MCP Client Compatibility: Why caniuse.dev Changes the Game

MCP Client Compatibility: Why caniuse.dev Changes the Game

AGNTCon + MCPCon North America 2026, AI, Application Development, DevSecOps, Events, Platform Engineering, PyTorch Conference /

MCPJam has launched caniuse.dev, a compatibility reference tool that surfaces behavioral differences in MCP support across AI clients including ChatGPT, Claude, Cursor, and GitHub Copilot. The tool addresses a critical gap between a shared protocol and consistent runtime behavior. For enterprise teams, it reframes MCP governance as a pipeline concern rather than a one-time audit.

MCP Client Compatibility: Why caniuse.dev Changes the Game Read More »

GenAI Cuts OSS Coding Time 50%—But Governance Lags | ECI Research

GenAI Cuts OSS Coding Time 50%—But Governance Lags | ECI Research

AGNTCon + MCPCon North America 2026, AI, Application Development, Automation, DevSecOps, Open Source, PyTorch Conference /

The Linux Foundation’s new study finds generative AI cuts open source coding time by up to 50%, but floods projects with low-quality AI-generated pull requests. With only 35% of projects holding a formal GenAI policy, governance has not kept pace with adoption. ECI Research breaks down what enterprise buyers and developer teams need to do about it.

GenAI Cuts OSS Coding Time 50%—But Governance Lags | ECI Research Read More »