PyTorch in Production: Boosting LLM Training and Inferencing on Ascend NPU
summary
Covers using PyTorch with Huawei Ascend NPUs to improve large-scale training and inference for LLMs by applying Ascend-optimized technologies and architecture changes in PyTorch. Describes implemented features and quality standards for heterogeneous compute and accelerator support, integration with ecosystem projects like DeepSpeed and vLLM, and presents real user case studies and best practices for production deployments on the Ascend platform.
AI-written summary · about
related talks
08mSponsored Session: Lightning Talk: PyTorch in Production: Boosting LLM Training and...
05mAscend & PyTorch: Pioneer New AI Open Ecosystem
29mMulti-Accelerator PyTorch Serving With NxD Inference and vLLM
13mAccelerating Model Training on Ascend Chips: An Industrial System for Profiling...
25mHow Modern PyTorch Supercharges Multimodal Training and Inference at Luma AI
28mBring the Power of ONNX to Spark as it Never Happened Before
07mPyTorch at Microsoft
10mvLLM & Deepspeed Updates