PyTorch Day Japan 2026 议程：用 ExecuTorch 把反向传播带进机器人实时推理

The non-obvious point is that the bottleneck in MoE training is often kernel launch overhead from Python-level expert loops, not the expert math itself. Engineers training MoE models in PyTorch can apply grouped execution to cut that overhead without restructuring their expert weights. The worked example uses biological foundation models, a niche domain where sparse expert routing is common.

官方账号PyTorch@PyTorch7 sourcesSource

The default switch to the rewritten NCCL backend and new NIXL one-sided transport affect anyone running distributed training, so tuning assumptions from earlier versions may need revisiting. Engineers building inference on Blackwell or Apple Silicon can check the live overview for concrete decode gains. Registration is open via the official link.

官方账号PyTorch@PyTorch11 sourcesSource