onnxsim/onnxsim

Health v2 · maintenance only

Adoption not part of the score

4390 stars · 432 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

ONNX Simplifier (onnxsim) is a tool that simplifies ONNX models by inferring the whole computation graph and replacing redundant operators with their constant outputs via constant folding. It is available as a Python package/CLI, a C++ tool, and a WebAssembly build that runs entirely in the browser.

Use cases

• simplify an onnx model exported from pytorch
• remove redundant operators from an onnx graph via constant folding
• fuse batchnorm into conv in an onnx model
• run shape inference on an onnx model
• verify a simplified onnx model produces the same outputs as the original
• optimize onnx models before deploying to mobile or edge runtimes
• inspect and simplify onnx models in the browser without uploading them

When to choose

• you exported a model to ONNX and the graph is more complicated than expected
• you want smaller, faster-loading ONNX models for inference deployment
• you need to fold constants or apply graph optimization passes to ONNX models
• you want a privacy-preserving, browser-based ONNX simplification tool

When to avoid

• you need to train or quantize models rather than simplify inference graphs
• you work with non-ONNX model formats like raw TensorFlow or PyTorch checkpoints
• you need a full model conversion framework rather than a simplifier

Facets

cli-tool · maturity active

machine-learning compiler developer-tools deep-learning machine-learning developer-tools python cpp cli browser wasm cross-platform onnx constant-folding graph-optimization shape-inference model-optimization pytorch

Member repositories

For agents

markdown · JSON · MCP: product_card(name="onnxsim/onnxsim")