#TPU

HeyGen ports Avatar IV model to Google Cloud TPUs

HeyGen has ported its 18B+ parameter Avatar IV video generation model to Google Cloud's Trillium (v6e) TPUs. Using torchax and XLA with FSDP and Ulysses sequence parallelism, the team achieved a 1.86x speedup for real-time streaming. The optimizations were deployed only after a two-tier quality gate confirmed pixel outputs stayed identical.

16:01Hardware

Google introduces autonomous LLM post-training on TPUs

The "autofinetune" project, detailed on the Google Developers Blog, offers an autonomous research loop that fully automates LLM post-training workflows. Developers define boundary conditions and evaluation metrics in a single Markdown file, and an AI agent iteratively edits training scripts, launches experiments, and commits verified hyperparameter optimizations to Git. The framework is built on G

15:46AI

#TPU

HeyGen ports Avatar IV model to Google Cloud TPUs

Google introduces autonomous LLM post-training on TPUs

Google to restrict access for free Gemini users

AWS introduces search agent fine-tuning with multi-turn RL on SageMaker AI

YouTube Shorts algorithm is changing: Recommendations will prioritize original content

DJI Osmo 360 II announced: What changed compared to the first generation?

OpenAI publishes a model guide for the GPT-6 family