Uber Michelangelo Is Now Open Source: Run a Battle-Tested ML Platform
Uber open sourced Michelangelo at KubeCon 2026, the ML platform behind 30M predictions per second. Here is how to run it on your ...
Google Open-Sources XLA Serving Engine: Serve JAX Models at 36x Throughput
Google open-sourced XLA Serving Engine (XSE), the C++ inference runtime powering YouTube Shorts and Gemini ...
Hugging Face Transformers MIC: Stop Silent ML Regressions
Hugging Face shipped Model Integration Contracts on Nov 3. Here is the UMT5 bug that ...
GKE Workspaces: Pause GPU Notebooks and Save State on GKE
Google's new GKE Workspaces lets you pause GPU notebooks and restore full state in seconds. ...
Tencent Hy4 Open Weights: Run a 770B Model With Computer Use
Tencent launched full Hy4 on November 4 — a 770B open-weight model with native computer ...
TorchTPU Is Open Source: Run PyTorch Natively on TPUs
Google and Meta open-sourced TorchTPU, the native PyTorch backend for TPUs. PyTorch/XLA goes read-only November 15. Here is what changes and how to ...
Kimi K3.1 Pro Open Weights: Nov 4 Benchmarks and API Guide
Kimi K3.1 Pro open weights land November 4. Here's what the benchmarks actually show, the ...
Mistral Large 4 Open Weights: Three Things to Fix Before Using It
Mistral Large 4 weights are live. Fix the reasoning effort default, check the real context ...
Mistral Large 4 Open Weight: Run a 1T Model Yourself
Mistral released the weights for Large 4 on October 29th — a 1 trillion-parameter MoE ...
Google TorchTPU: PyTorch on TPUs, XLA Gets Archived
Google and Meta open-sourced TorchTPU, making TPU a native PyTorch device. PyTorch/XLA is archived November ...