When should I use TorchScript tracing versus scripting?

Use tracing for models with static execution paths as it is faster; use scripting for models containing dynamic Python control flow like if-else statements or loops.

Does this skill support ONNX Runtime?

Yes, it includes patterns for exporting models to the ONNX format and configuring sessions for cross-platform acceleration via ONNX Runtime.

Can this skill help with mobile deployment?

Yes, it provides specific workflows for the PyTorch Mobile Lite Interpreter and optimizing models for resource-constrained environments like iOS and Android.

How does quantization improve model performance?

Quantization reduces weight precision from float32 to int8, typically shrinking model size by 4x and significantly speeding up CPU inference by 2-3x.

Is LibTorch integration covered?

Absolutely, the skill includes C++ implementation patterns for loading and executing serialized TorchScript models in native applications.

PyTorch Model Deployment & Optimization

Name: PyTorch Model Deployment & Optimization
Author: tondevrel

bytondevrel

•

데이터 과학 및 ML

Deploys and optimizes PyTorch models for production environments, edge devices, and high-performance C++ applications.

This skill empowers Claude to guide developers through the complex process of moving machine learning models from research notebooks to production-ready systems. It provides expert patterns for model serialization using TorchScript (both Tracing and Scripting), cross-platform ONNX exports, and integration with the LibTorch C++ API. Furthermore, it offers specialized strategies for inference optimization, including post-training quantization and pruning, ensuring models are lightweight enough for mobile and edge hardware while maintaining high performance without sacrificing accuracy.

주요 기능

01TorchScript serialization via JIT tracing and scripting for Python-free execution.

02Mobile-specific deployment workflows for iOS, Android, and edge hardware like NVIDIA Jetson.

031 GitHub stars

04LibTorch integration patterns for performance-critical C++ inference engines.

05Model optimization via post-training quantization (INT8) and pruning strategies.

06Seamless ONNX export with dynamic axis configuration for cross-platform compatibility.

사용 사례

01Embedding PyTorch models into high-throughput C++ software environments.

02Optimizing neural networks for real-time inference on mobile devices and IoT hardware.

03Transitioning research models into production-grade FastAPI, Go, or Rust backends.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add tondevrel/scientific-agent-skills pytorch-deployment

For use in Claude.ai and ChatGPT

Download Skill