v0.2.1

What's Changed

• feat: add --source torch-export option to CoreML backend by @Boulaouaney in #9

Full Changelog: v0.2.0...v0.2.1

v0.2.0

What's Changed

• feat: add onnxslim optimization to ONNX export by @Boulaouaney in #8

Full Changelog: v0.1.0...v0.2.0

v0.1.0

Initial release of timmx — an extensible CLI and Python package for exporting timm models to deployment formats.

Backends

• ONNX — dynamo-based export with fallback, dynamic batch, opset selection
• Core ML — mlprogram/neuralnetwork, dynamic batch, compute precision control
• LiteRT / TFLite — fp32, fp16, dynamic-int8, int8 quantization with calibration data, NHWC input layout
• ncnn — export via pnnx with automatic cleanup of intermediate files
• TensorRT — fp32, fp16, int8 with calibration, dynamic batch, requires CUDA
• ExecuTorch — XNNPack and CoreML delegates, fp32 and int8 (PT2E quantization), dynamic batch
• torch.export — standard PyTorch export with dynamic shapes
• TorchScript — trace and script modes

Usage

pip install timmx
pip install 'timmx[onnx]'  # install backend extras as needed

timmx export onnx resnet18 --pretrained --output model.onnx
timmx doctor  # check which backends are available

Requirements

• Python >= 3.11
• PyTorch >= 2.5.0