source: YOLO26 @320/@256 CoreML recipe is static (batch 1)

Corresponding Source for the re-exported YOLO26 @320/@256 CoreML packages: build-camstack-models.py and source/README.md, verbatim from the CamStack repository.

source/README.md

@@ -39,7 +39,7 @@ What the recipes encode (the hosted files record each of these):
 | Folder | Files | ONNX | CoreML | OpenVINO IR |
 | --- | --- | --- | --- | --- |
 | `objectDetection/yolo26` | `camstack-yolo26{n,s,m,l,x}` @640 | static, opset 17 | static, batch 1 | FP16 as `<stem>.xml` and `<stem>-fp16.xml`; INT8 `<stem>-int8.xml` (128 images) |
-| `objectDetection/yolo26` | `camstack-yolo26{n,s,m,l}-{320,256}` | dynamic axes, opset 17 | flexible batch 1..16, FP16 | FP32 `<stem>.xml`; INT8 `<stem>-int8.xml` (128 images) |
+| `objectDetection/yolo26` | `camstack-yolo26{n,s,m,l}-{320,256}` | dynamic axes, opset 17 | static, batch 1, FP16, as `<stem>-static.mlpackage` and `<stem>.mlpackage` | FP32 `<stem>.xml`; INT8 `<stem>-int8.xml` (128 images) |
 | `objectDetection/yolov9` | `camstack-yolov9{t,s,m,c}-{320,640}` | static, opset 13 | static, batch 1 | FP16 `<stem>-fp16.xml`; INT8 `<stem>-int8.xml` (300 images) |
 | `segmentation/yolo26-seg` | `camstack-yolo26{n,s,m}-seg` @640 | static, opset 17 | static, batch 1 | FP16 `<stem>.xml` |
 
@@ -48,9 +48,15 @@ YOLO26 @640 and YOLO26-seg 8.4.24 / 2.10.0 / 9.0 / 2026.2.1; YOLO26 @320 and
 @256 8.4.92 / 2.13.0 for ONNX, 2.5.1 for CoreML / 8.3.0 / 2026.2.1; YOLOv9
 8.4.98 / 2.13.0 / 9.0 / 2026.2.1. The script warns when the installed Ultralytics differs.
 With these versions a rebuild gives the hosted weights. Other versions give the
-same weights in a possibly different graph. The flexible-batch CoreML exports
-need coremltools 8.3 with PyTorch 2.5.1, the macOS pins in `requirements.txt`;
-coremltools 9 aborts on them.
+same weights in a possibly different graph. The @320 and @256 CoreML exports
+use coremltools 8.3 with PyTorch 2.5.1, the macOS pins in `requirements.txt`.
+
+The @320 and @256 CoreML packages were batch-flexible (1..16) until
+2026-10-07. That package aborts any process that loads it on macOS 26.2, on
+every compute unit, because Core ML's `gather_nd` shape pass fails on a
+flexible batch. They are now exported with a static batch of 1. The same
+package is published as `<stem>-static.mlpackage` (the name the catalog uses)
+and under the old `<stem>.mlpackage` name.
 
 ### INT8 calibration
 

source/build-camstack-models.py

@@ -134,6 +134,22 @@ class Recipe:
     openvino: tuple[OpenvinoOutput, ...] = ()
     calib_samples: int = 0
     note: str = ''
+    # The CoreML package's stem when it differs from `name`. A package that had
+    # to be re-exported in an incompatible way gets a NEW file name, because a
+    # node keeps any package already on disk under the old one forever.
+    coreml_name: Optional[str] = None
+    # Earlier stems the SAME package is also published under, so a release
+    # whose catalog still names them downloads the working package too.
+    coreml_aliases: tuple[str, ...] = ()
+
+    @property
+    def coreml_stem(self) -> str:
+        return self.coreml_name or self.name
+
+    def coreml_paths(self) -> list[str]:
+        """Every mirror path of the CoreML package: its own, then its aliases."""
+        return [f'{self.folder}/coreml/{stem}.mlpackage/'
+                for stem in (self.coreml_stem, *self.coreml_aliases)]
 
     def outputs(self) -> list[str]:
         """Mirror paths this recipe produces (an .mlpackage is a directory)."""
@@ -141,7 +157,7 @@ class Recipe:
         if self.onnx is not None:
             out.append(f'{self.folder}/onnx/{self.name}.onnx')
         if self.coreml is not None:
-            out.append(f'{self.folder}/coreml/{self.name}.mlpackage/')
+            out.extend(self.coreml_paths())
         for ov in self.openvino:
             stem = f'{self.folder}/openvino/{self.name}{ov.suffix}'
             out.extend([f'{stem}.xml', f'{stem}.bin'])
@@ -176,14 +192,27 @@ def _yolo26_full(tier: str) -> Recipe:
 
 
 def _yolo26_reduced(tier: str, res: int) -> Recipe:
-    """yolo26<tier>-<res>: dynamic-axes ONNX, batch-flexible (1..16) FP16 CoreML, IR fp32 + INT8."""
+    """yolo26<tier>-<res>: dynamic-axes ONNX, STATIC FP16 CoreML, IR fp32 + INT8.
+
+    The CoreML package was batch-flexible (1..16) until 2026-10-07, and that
+    package aborts the process that loads it (SIGABRT, macOS 26.2, every compute
+    unit, coremltools 8.3 and 9.0 alike): Core ML's Espresso shape pass frees an
+    unallocated pointer in `gather_nd` when the input batch is a range. YOLO26's
+    NMS-free head is the only hosted graph with a `gather_nd`, which is why the
+    one other batch-flexible package (bird-classifier) never hit it, and why the
+    static @640 YOLO26 packages load. The static export loads
+    and runs on every compute unit, so it is published under a new stem
+    (`-static`) and under the old one.
+    """
     return Recipe(
         name=f'camstack-yolo26{tier}-{res}', folder='objectDetection/yolo26', imgsz=res,
         toolchain=TC_2607_REDUCED, checkpoint=f'yolo26{tier}',
         onnx=OnnxExport(dynamic=True, opset=17),
-        coreml=CoremlExport(dynamic=True, batch=16, half=True),
+        coreml=CoremlExport(dynamic=False, batch=1, half=True),
         openvino=(OpenvinoOutput('', 'fp32'), OpenvinoOutput('-int8', 'int8')),
         calib_samples=128,
+        coreml_name=f'camstack-yolo26{tier}-{res}-static',
+        coreml_aliases=(f'camstack-yolo26{tier}-{res}',),
     )
 
 
@@ -354,7 +383,9 @@ def build_coreml(recipe: Recipe, out: Path, verify: bool) -> Path:
         checkpoint_path(recipe), format='coreml', imgsz=recipe.imgsz,
         dynamic=c.dynamic, batch=c.batch, half=c.half, nms=False,
     )
-    dest = place(src, out / recipe.folder / 'coreml' / f'{recipe.name}.mlpackage')
+    dest = place(src, out / recipe.folder / 'coreml' / f'{recipe.coreml_stem}.mlpackage')
+    for alias in recipe.coreml_aliases:
+        place(dest, out / recipe.folder / 'coreml' / f'{alias}.mlpackage')
     if verify and c.dynamic:
         verify_coreml_batch(dest, recipe.imgsz)
     return dest