onnxonnxOfficialยท1 Agent Skills Included

ir-py

Build and transform ONNX model graphs in memory

Builds, analyzes, and transforms ONNX machine learning models directly in memory with full spec support. Eliminates slow protobuf handling, tensor size limits, and fragile graph edits through zero-copy tensor access and safe mutation. Guides the creation of reusable transformation passes for tasks like dead code removal, constant folding, and graph normalization.
npx skills add onnx/ir-py --all -g -y

All Skills in This Repository (1)

Pure Emerald Level Indicators

Frequently Asked Questions

FAQPage Schema
How to install ir-py?โ–ผ

Run `npx skills add onnx/ir-py --all -g -y` in your terminal to install all skills in this repository globally.

What is ONNX IR used for?โ–ผ

It is a Python library for building, analyzing, and transforming ONNX model graphs in memory without depending on protobuf after loading.

How do I write an ONNX transformation pass?โ–ผ

The pass-writing skill teaches your agent the correct base classes, traversal patterns, and safety rules to create reliable graph transformation passes.

Does ONNX IR handle large model tensors?โ–ผ

Yes. It uses memory-mapped external tensors and zero-copy access, so there is no tensor size limitation and memory usage stays low.

Can I modify a graph while iterating over it?โ–ผ

Yes. The library's iterators are designed for safe mutation during traversal, and the skill explains forward and reverse iteration patterns.

Related Repositories in Software Engineering

View All in Software Engineeringโ†’