Sponsored Session: Lightning Talk: PyTorch in Production: Boosting LLM Training and...
summary
Shows concrete PyTorch changes for Ascend NPU: adds privateuse1 plugin loading in PyTorch 2.1 so backends auto-import via Python entry points, provides an OpenRidge CPU-sim backend for compatibility tests (more than 10 capability issues found in two months), implements an MPU graph to cut inference latency from >100 ms to 28 ms, inductor+Triton backend in PyTorch 2.6, and operator/runtime tweaks that claim up to 50% CPU overhead reduction and production-grade training and inference on Ascend.
AI-written summary · about
related talks
11mPyTorch in Production: Boosting LLM Training and Inferencing on Ascend NPU
14mLightning Talk: The Fastest Path to Production: PyTorch Inference in Python
54mKeynote: PyTorch 2.1 Technical Deep Dive
08mLightning Talk: Enhancements Made to MPS Backend in PyTorch for Applications Running...
12mLightning Talk: State of PyTorch
06mKeynote: PyTorch Lightning: Powering the GenAI Revolution from Research to the...
16mLightning Talk: Efficient Inference at the Edge: Performance You Need at the Lowest...
14mLightning Talk: Every Millisecond Counts: The Fine-tuning Journey of an Ultra-Eff...