Google open-sourced TorchTPU, a native PyTorch backend for Cloud TPUs that reaches 52% MFU training Qwen 3 8B.
DeepSeek open-sourced DeepInfer, its production inference framework for DeepSeek-V4.1-Flash, reporting over 25,000 decode tokens per second.
OpenAI has reported significant gains in coding across 20 programming languages and migrating legacy software.
A 10-step guide to on-device AI, covering model selection, quantization, deployment runtimes, hardware acceleration, hybrid ...