OSSxaskasdf/ntransformer — Llama 70B on Single RTX 3090 via NVMe-to-GPU BypassGitHub·high signalC++ CUDA inference engine streams model layers through GPU via PCIe. Optional NVMe-to-GPU bypass. Show HN 380 pts.SourceSource pageGitHub↳ Follow the threadStack layerPattern: Terminal-Native, Bring-Your-Own-Model Coding Agents Are Proliferating and SpecializingGitHubStack layerPattern: Token Cost Has Become Its Own Tooling CategoryGitHubStack layerCode-first analysis: make the agent write a script that returns only results instead of reading files into contextGitHubStack layer / Threat patternMeta Stands Up a Cloud Business to Sell AI Compute — CoreWeave Stock Drops 14%The Motley FoolStack layer / Update threadJim Fan: NVIDIA Open-Sources ASPIRE, and 'the Training Output Changes From Model Weights to a Continuously Expanding Skill Library'BigGo FinancePolicy dependency / Stack layerauthsome Gives AI Agents Credentials Without Ever Exposing Your SecretsGitHubStack layer / Update threadZML Releases LLMD Alpha — a Zig-Based, Chip-Agnostic Inference Server Across 5 ArchitecturesTechCrunch / ZMLStack layer / Threat patternCross-model 'prompt laundering' and GuardFall shell-injection flaws hit major coding assistantsThreat-Modeling.com