Does this skill support GPU acceleration?

Yes, it includes specific constraints and design patterns for utilizing CUDA and Metal via frameworks like candle and tch-rs, including async data loading to keep the GPU busy.

How should I handle model loading in a web server context?

It implements the singleton pattern using OnceLock, ensuring that heavy ML models are loaded into memory only once and reused across all incoming requests.

Which Rust ML framework is best for inference-only projects?

For lightweight and portable inference, the skill recommends using 'tract' with ONNX models as it is highly optimized for production environments.

How do I prevent memory issues when handling large tensors in Rust?

The skill provides patterns for using ndarray views and zero-copy operations to avoid cloning large datasets, which is the primary cause of memory bandwidth bottlenecks.

Rust Machine Learning & AI

Name: Rust Machine Learning & AI
Author: lywa1998

bylywa1998

数据科学与机器学习

Guides the development of high-performance ML and AI applications in Rust using memory-efficient patterns and GPU acceleration.

关于

The domain-ml skill provides specialized expertise for building production-grade machine learning systems within the Rust ecosystem. It focuses on solving common domain challenges like memory bandwidth bottlenecks, GPU utilization, and model portability. By providing implementation patterns for frameworks such as Candle, Burn, and Tract, the skill ensures developers follow best practices for zero-copy tensor operations, singleton model loading, and efficient batch processing to maximize performance and reliability in AI-driven applications.

主要功能

0 GitHub stars
Framework selection guidance for Tract, Candle, Burn, and tch-rs
Production-ready patterns for singleton model loading and batch inference
Memory-efficient design patterns using zero-copy and tensor views
Standardized ONNX model integration and portability workflows
GPU acceleration strategies for CUDA and Metal backends

使用场景

Building high-throughput inference servers for ONNX-based models
Developing native Rust neural networks with GPU-accelerated training
Creating low-latency data preprocessing pipelines for large-scale datasets

关于

主要功能

0 GitHub stars
Framework selection guidance for Tract, Candle, Burn, and tch-rs
Production-ready patterns for singleton model loading and batch inference
Memory-efficient design patterns using zero-copy and tensor views
Standardized ONNX model integration and portability workflows
GPU acceleration strategies for CUDA and Metal backends

使用场景

Building high-throughput inference servers for ONNX-based models
Developing native Rust neural networks with GPU-accelerated training
Creating low-latency data preprocessing pipelines for large-scale datasets