A preparation workflow for running machine-learning models on NPUs, specialized processors for neural-network workloads. It covers understanding the code, setting up the CANN environment, preparing test data, collecting profiling data, and checking numerical accuracy.
A performance-bottleneck analysis skill that combines profiling data with source-code structure analysis. Profiling records where a program spends time or resources.
A Chinese-language implementation guide for performance optimisation using four approaches: deduplication, reuse, masking, and replacement. It covers changes such as combining operations, preallocating buffers, and using compilation tools.
A Chinese-language skill for checking and debugging numerical accuracy, including comparisons between GPU and NPU results. An NPU is a processor designed for neural-network workloads.
A set of engineering practices for planning folder structures, managing versions, and maintaining documentation. It also covers organising code directories and managing Git branches and commits.
An exploratory workflow for improving inference performance on neural processing units, or NPUs—specialised hardware for running AI models. It covers collecting data, analysing results, trying optimisations, checking performance, and recording findings without requiring a fixed framework.
A workflow for adapting machine-learning models to run on NPUs, specialised chips for neural-network workloads. It covers the environment, model weights, inference code, smoke tests, accuracy checks, and out-of-memory diagnosis.
A guide for deciding when and how to split machine-learning inference across multiple NPUs, specialised chips for neural-network workloads.
★not rated 5▲
+1 1mo agoA78 tokens
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: