feat(ml): coreml (#17718)

* coreml

* add test

* use arena by default in native installation

* fix tests

* add env to docs

* remove availability envs

machine-learning/immich_ml/config.py

@@ -61,6 +61,7 @@ class Settings(BaseSettings):
     request_threads: int = os.cpu_count() or 4
     model_inter_op_threads: int = 0
     model_intra_op_threads: int = 0
+    model_arena: bool = True
     ann: bool = True
     ann_fp16_turbo: bool = False
     ann_tuning_level: int = 2

machine-learning/immich_ml/models/constants.py

@@ -79,6 +79,7 @@ SUPPORTED_PROVIDERS = [
     "CUDAExecutionProvider",
     "ROCMExecutionProvider",
     "OpenVINOExecutionProvider",
+    "CoreMLExecutionProvider",
     "CPUExecutionProvider",
 ]
 

machine-learning/immich_ml/sessions/ort.py

@@ -96,6 +96,14 @@ class OrtSession:
                         "precision": "FP32",
                         "cache_dir": (self.model_path.parent / "openvino").as_posix(),
                     }
+                case "CoreMLExecutionProvider":
+                    options = {
+                        "ModelFormat": "MLProgram",
+                        "MLComputeUnits": "ALL",
+                        "SpecializationStrategy": "FastPrediction",
+                        "AllowLowPrecisionAccumulationOnGPU": "1",
+                        "ModelCacheDirectory": (self.model_path.parent / "coreml").as_posix(),
+                    }
                 case _:
                     options = {}
             provider_options.append(options)
@@ -115,7 +123,7 @@ class OrtSession:
     @property
     def _sess_options_default(self) -> ort.SessionOptions:
         sess_options = ort.SessionOptions()
-        sess_options.enable_cpu_mem_arena = False
+        sess_options.enable_cpu_mem_arena = settings.model_arena
 
         # avoid thread contention between models
         if settings.model_inter_op_threads > 0: