66.55% mIoU on VOC2012 val (200 images) against 66.56% for the XNNPACK int8
file, at 3.20 ms/inference on an S26 Ultra. Returns a class index per pixel,
(1,520,520) int32, instead of (1,21,520,520) logits.
Core ML: emit a (1,520,520) int32 index map instead of (1,21,520,520) logits
The argmax now runs inside the graph, built from reduce_max and elementwise ops so it stays on the ANE. 1.05x to 1.63x faster and 7 to 48 percent lower peak memory on an iPhone 16. XNNPACK variants are unchanged.
Core ML: emit a (1,520,520) int32 index map instead of (1,21,520,520) logits
The argmax now runs inside the graph, built from reduce_max and elementwise ops so it stays on the ANE. 1.05x to 1.63x faster and 7 to 48 percent lower peak memory on an iPhone 16. XNNPACK variants are unchanged.
Core ML: emit a (1,520,520) int32 index map instead of (1,21,520,520) logits
The argmax now runs inside the graph, built from reduce_max and elementwise ops so it stays on the ANE. 1.05x to 1.63x faster and 7 to 48 percent lower peak memory on an iPhone 16. XNNPACK variants are unchanged.
Core ML: emit a (1,520,520) int32 index map instead of (1,21,520,520) logits
The argmax now runs inside the graph, built from reduce_max and elementwise ops so it stays on the ANE. 1.05x to 1.63x faster and 7 to 48 percent lower peak memory on an iPhone 16. XNNPACK variants are unchanged.
quantized duplicated precisions.json, which is authoritative for the
quantized/non-quantized partition. default expressed a per-repository
recommendation nothing consumed: the library resolves its own per-platform
default. Both are gone from the schema.
Adds library_name to the model card, and pipeline_tag where it was missing.
Brings the config and card licenses into agreement, and repairs the variant
declarations that did not match the files actually published here.
Core ML: drop the fp32 variant, fp16 is equivalent in task terms
fp32 is both slower and larger than fp16 on device, and the ANE will not accept fp32 at all, so an fp32 Core ML build is structurally excluded from the accelerator that makes Core ML worth using.
Equivalence was gated with a task metric rather than raw tensor distance: pixel argmax agreement 99.88% across 8 real images. Raw relative L2 on logits reads ~1% for the segmentation models, but that is logit noise which does not change the predicted class per pixel.
Models where fp16 and fp32 genuinely diverge (ssdlite320 86% box agreement, fast-sam 93%) keep both variants.
v0.9.0 and below are untouched.
Remove MLX artifacts: slower than Core ML on device
MLX measured slower than Core ML on an iPhone 16 for this model, at matched fp32 precision and against the shipped Core ML fp16 build. These exporters only ever declared an MLX fp32 variant, so fp32-vs-fp32 is the fair and conservative comparison, and MLX still loses it.
The .pte and config.json are removed; each mlx/ dir keeps a NOTES.md with this model's own measurements. The exporter is retained in export-scripts so this can be re-tested against a future ExecuTorch/MLX release.
MLX for these models was only ever published on main / v0.10.0; no tag at v0.9.0 or below carried it. v0.9.0 and below are untouched.
