Import/Export Compatibility Matrix¶

This page is generated from docs/import_export_compatibility_matrix.json.

Schema version: interop_import_export_matrix_v1
Snapshot generated at (UTC): 2026-04-14T00:00:00+00:00
Source of truth: docs/import_export_compatibility_matrix.json
Generated by: python scripts/generate_import_export_matrix_docs.py --refresh-runtime-snapshot

Bridge	Runtime Available	Import	Export	DTypes	Layout	Shape Constraints	Zero-Copy	Fallback Reason Codes
lightning	Yes	numpy_ndarray	numpy_ndarray	float32 int64(token_ids)	seq_dmodel_2d nchw_4d	ops-specific (validated at runtime)	no	none
torch	No	torch.Tensor	torch.Tensor	float32 int64(token_ids)	contiguous tensor preferred	runner/adapter contract validated	eligible on cpu+contiguous+dtype-match	none interop_torch_tensor_boundary_copy
tensorflow	No	tf.Tensor or numpy shim	tf.Tensor or numpy shim	float32 int64(token_ids)	tensor/ndarray contiguous preferred	runner/adapter contract validated	no	none tf_runtime_unavailable tf_tensor_boundary_copy tf_runner_graph_policy_forced_eager tf_runner_graph_execute_failed tf_runner_interop_policy_forced_lightning tf_runner_unknown_fallback
coreml	Yes	runner input + optional .mlmodel/.mlmodelc path	runner output numpy + coreml benchmark report	float32 int64(token_ids)->runner preprocess	seq_dmodel_2d	alpha adapter; benchmark bridge path	eligible on float32 contiguous inputs	none coreml_runtime_unavailable coreml_model_path_missing coreml_inference_failed coreml_tensor_boundary_copy coreml_runner_graph_policy_forced_eager coreml_runner_unknown_fallback
mlx	No	mlx.core array (or numpy shim)	numpy (and optional mlx bridge path)	float32 int64(token_ids)->runner preprocess	seq_dmodel_2d	alpha adapter; runtime optional	eligible on float32 contiguous inputs	none mlx_runtime_unavailable mlx_tensor_boundary_copy mlx_bridge_execute_failed mlx_runner_graph_policy_forced_eager mlx_runner_unknown_fallback