deftruth/optimizer

ONNX Optimizer is a tool that takes AI models saved in the ONNX format and makes them smaller, faster, and more efficient. ONNX is a common file format that lets AI models run across different software environments, but models exported to this format often carry redundant steps or bloated operations. This tool cleans them up so they run better in production. At a high level, it works by applying "optimization passes" to a model file. A pass is essentially a rule that looks for known inefficiencies and rewrites them. For example, it can fuse multiple operations into a single step, or eliminate parts of the model that do not contribute to the final output. You can run it as a simple command-line tool, pointing it at an input model and getting a slimmer output model back. You can let it apply a default set of passes, or pick specific ones. The people who benefit most from this are engineers and teams deploying AI models to real users. If you trained a model in PyTorch or TensorFlow and exported it to ONNX so it can run on a phone, in a browser, or on a server, running it through this optimizer first can reduce load times and inference costs. It is especially useful for teams building their own ONNX-based serving infrastructure, since it bundles common optimization techniques into a reusable tool rather than requiring each team to write their own. The project is straightforward to use, you can install it with a single command and run it without writing any code. However, the README does not go into much detail on the specific optimization passes available or what each one does. You would need to run it with the flag that prints all available passes to see the full list and experiment from there.