Roadmap¶
Planned¶
- Broaden coverage of JAX, Flax NNX/Linen, and Equinox components.
- Expand SotA example support for vision and language models.
- Improve support for physics-based simulations.
- Support for MaxDiffusion.
Current Version¶
jax2onnx 0.12.0 - Layout controls, opset 23 defaults, and regression hardening¶
-
NCHW boundary layout support: Added
inputs_as_nchw/outputs_as_nchwforto_onnx(...)andallclose(...), with layout-optimization docs/tests and transpose-cleanup improvements for Conv-heavy graphs (PR #164, #172). -
Depth-to-space and residual-stack coverage: Added
dm_pix.depth_to_spacelowering to ONNXDepthToSpaceand expanded NNX regression examples/tests for depth-to-space and nested residual groups (PR #167, #175, #176). -
Primitive and IR improvements: Added
jax.numpy.meanlowering toReduceMean; fixed symbolicdim_as_valuehandling; and stabilized dynamic reshape folding used by CLIP/MaxText exports (PR #170, #171, #179). -
ONNX opset 23 path for attention models: Added opset >= 23 RotaryEmbedding/Attention support and made opset 23 the default in
to_onnx(...)(PR #177). -
Gather/scatter regression fixes: Fixed scatter-add broadcast window handling and issue #52 lowering edge cases; fixed gather indexing and
vmap(dynamic_slice_in_dim)gather lowering regressions (PR #181, #183, #184). -
Compatibility refresh: Expanded tested Python versions to 3.11-3.14 and updated runtime dependency floors (
onnx,onnxruntime,dm-pix) for the new paths.
Past Versions¶
See Past Versions for the full release archive.