fxn 0.0.50__tar.gz → 0.0.52__tar.gz

{fxn-0.0.50 → fxn-0.0.52}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: fxn
-Version: 0.0.50
+Version: 0.0.52
 Summary: Run prediction functions locally in Python. Register at https://fxn.ai.
 Author-email: "NatML Inc." <hi@fxn.ai>
 License:                                  Apache License

{fxn-0.0.50 → fxn-0.0.52}/fxn/beta/__init__.py

@@ -5,6 +5,7 @@
 
 from .metadata import (
     CoreMLInferenceMetadata, LiteRTInferenceMetadata, LlamaCppInferenceMetadata,
-    ONNXInferenceMetadata, ONNXRuntimeInferenceSessionMetadata
+    ONNXInferenceMetadata, ONNXRuntimeInferenceSessionMetadata, OpenVINOInferenceMetadata,
+    QnnInferenceMetadata
 )
 from .remote import RemoteAcceleration

{fxn-0.0.50 → fxn-0.0.52}/fxn/beta/metadata.py

@@ -30,8 +30,8 @@ class CoreMLInferenceMetadata (BaseModel):
     Metadata required to lower a PyTorch model for inference on iOS, macOS, and visionOS with CoreML.
     """
     kind: Literal["meta.inference.coreml"] = "meta.inference.coreml"
-    model: Annotated[object, BeforeValidator(_validate_torch_module)] = Field(description="PyTorch module to apply metadata to.")
-    model_args: list[object] = Field(description="Positional inputs to the model.")
+    model: Annotated[object, BeforeValidator(_validate_torch_module)] = Field(description="PyTorch module to apply metadata to.", exclude=True)
+    model_args: list[object] = Field(description="Positional inputs to the model.", exclude=True)
     model_config = ConfigDict(arbitrary_types_allowed=True, frozen=True)
 
 class ONNXInferenceMetadata (BaseModel):
@@ -39,8 +39,8 @@ class ONNXInferenceMetadata (BaseModel):
     Metadata required to lower a PyTorch model for inference.
     """
     kind: Literal["meta.inference.onnx"] = "meta.inference.onnx"
-    model: Annotated[object, BeforeValidator(_validate_torch_module)] = Field(description="PyTorch module to apply metadata to.")
-    model_args: list[object] = Field(description="Positional inputs to the model.")
+    model: Annotated[object, BeforeValidator(_validate_torch_module)] = Field(description="PyTorch module to apply metadata to.", exclude=True)
+    model_args: list[object] = Field(description="Positional inputs to the model.", exclude=True)
     model_config = ConfigDict(arbitrary_types_allowed=True, frozen=True)
 
 class ONNXRuntimeInferenceSessionMetadata (BaseModel):
@@ -48,8 +48,8 @@ class ONNXRuntimeInferenceSessionMetadata (BaseModel):
     Metadata required to lower an ONNXRuntime `InferenceSession` for inference.
     """
     kind: Literal["meta.inference.onnxruntime"] = "meta.inference.onnxruntime"
-    session: Annotated[object, BeforeValidator(_validate_ort_inference_session)] = Field(description="ONNXRuntime inference session to apply metadata to.")
-    model_path: Path = Field(description="ONNX model path. The model must exist at this path in the compiler sandbox.")
+    session: Annotated[object, BeforeValidator(_validate_ort_inference_session)] = Field(description="ONNXRuntime inference session to apply metadata to.", exclude=True)
+    model_path: Path = Field(description="ONNX model path. The model must exist at this path in the compiler sandbox.", exclude=True)
     model_config = ConfigDict(arbitrary_types_allowed=True, frozen=True)
 
 class LiteRTInferenceMetadata (BaseModel):
@@ -57,8 +57,27 @@ class LiteRTInferenceMetadata (BaseModel):
     Metadata required to lower PyTorch model for inference with LiteRT (fka TensorFlow Lite).
     """
     kind: Literal["meta.inference.litert"] = "meta.inference.litert"
-    model: Annotated[object, BeforeValidator(_validate_torch_module)] = Field(description="PyTorch module to apply metadata to.")
-    model_args: list[object] = Field(description="Positional inputs to the model.")
+    model: Annotated[object, BeforeValidator(_validate_torch_module)] = Field(description="PyTorch module to apply metadata to.", exclude=True)
+    model_args: list[object] = Field(description="Positional inputs to the model.", exclude=True)
+    model_config = ConfigDict(arbitrary_types_allowed=True, frozen=True)
+
+class OpenVINOInferenceMetadata (BaseModel):
+    """
+    Metadata required to lower PyTorch model for interence with Intel OpenVINO.
+    """
+    kind: Literal["meta.inference.openvino"] = "meta.inference.openvino"
+    model: Annotated[object, BeforeValidator(_validate_torch_module)] = Field(description="PyTorch module to apply metadata to.", exclude=True)
+    model_args: list[object] = Field(description="Positional inputs to the model.", exclude=True)
+    model_config = ConfigDict(arbitrary_types_allowed=True, frozen=True)
+
+class QnnInferenceMetadata (BaseModel):
+    """
+    Metadata required to lower a PyTorch model for inference on Qualcomm accelerators with QNN SDK.
+    """
+    kind: Literal["meta.inference.qnn"] = "meta.inference.qnn"
+    model: Annotated[object, BeforeValidator(_validate_torch_module)] = Field(description="PyTorch module to apply metadata to.", exclude=True)
+    model_args: list[object] = Field(description="Positional inputs to the model.", exclude=True)
+    backend: Literal["cpu", "gpu"] = Field(default="cpu", description="QNN backend to execute the model.", exclude=True) # CHECK # Add `htp`
     model_config = ConfigDict(arbitrary_types_allowed=True, frozen=True)
 
 class LlamaCppInferenceMetadata (BaseModel): # INCOMPLETE
@@ -66,5 +85,5 @@ class LlamaCppInferenceMetadata (BaseModel): # INCOMPLETE
     Metadata required to lower a GGUF model for LLM inference.
     """
     kind: Literal["meta.inference.gguf"] = "meta.inference.gguf"
-    model_path: Path = Field(description="GGUF model path. The model must exist at this path in the compiler sandbox.")
+    model_path: Path = Field(description="GGUF model path. The model must exist at this path in the compiler sandbox.", exclude=True)
     model_config = ConfigDict(arbitrary_types_allowed=True, frozen=True)

{fxn-0.0.50 → fxn-0.0.52}/fxn/compile.py

@@ -9,25 +9,39 @@ from inspect import isasyncgenfunction, iscoroutinefunction
 from pathlib import Path
 from pydantic import BaseModel, ConfigDict, Field
 from types import ModuleType
-from typing import Literal
+from typing import Any, Callable, Literal, ParamSpec, TypeVar, cast
 
 from .beta import (
     CoreMLInferenceMetadata, LiteRTInferenceMetadata, LlamaCppInferenceMetadata,
-    ONNXInferenceMetadata, ONNXRuntimeInferenceSessionMetadata
+    ONNXInferenceMetadata, ONNXRuntimeInferenceSessionMetadata, OpenVINOInferenceMetadata,
+    QnnInferenceMetadata
 )
 from .sandbox import Sandbox
 from .types import AccessMode
 
-CompileTarget = Literal["android", "ios", "linux", "macos", "visionos", "wasm", "windows"]
+CompileTarget = Literal[
+    "android",
+    "ios",
+    "linux",
+    "macos",
+    "visionos",
+    "wasm",
+    "windows"
+]
 
 CompileMetadata = (
     CoreMLInferenceMetadata             |
     LiteRTInferenceMetadata             |
     LlamaCppInferenceMetadata           |
     ONNXInferenceMetadata               |
-    ONNXRuntimeInferenceSessionMetadata
+    ONNXRuntimeInferenceSessionMetadata |
+    OpenVINOInferenceMetadata           |
+    QnnInferenceMetadata
 )
 
+P = ParamSpec("P")
+R = TypeVar("R")
+
 class PredictorSpec (BaseModel):
     """
     Descriptor of a predictor to be compiled.
@@ -36,6 +50,7 @@ class PredictorSpec (BaseModel):
     description: str = Field(description="Predictor description. MUST be less than 100 characters long.", min_length=4, max_length=100)
     sandbox: Sandbox = Field(description="Sandbox to compile the function.")
     targets: list[str] | None = Field(description="Targets to compile this predictor for. Pass `None` to compile for our default targets.")
+    metadata: list[object] = Field(default=[], description="Metadata to use while compiling the function.")
     access: AccessMode = Field(description="Predictor access.")
     card: str | None = Field(default=None, description="Predictor card (markdown).")
     media: str | None = Field(default=None, description="Predictor media URL.")
@@ -49,13 +64,13 @@ def compile (
     sandbox: Sandbox=None,
     trace_modules: list[ModuleType]=[],
     targets: list[CompileTarget]=None,
-    access: AccessMode=AccessMode.Private,
     metadata: list[CompileMetadata]=[],
+    access: AccessMode=AccessMode.Private,
     card: str | Path=None,
     media: Path=None,
     license: str=None,
     **kwargs
-):
+) -> Callable[[Callable[P, R]], Callable[P, R]]:
     """
     Create a predictor by compiling a stateless function.
 
@@ -65,8 +80,8 @@ def compile (
         sandbox (Sandbox): Sandbox to compile the function.
         trace_modules (list): Modules to trace and compile.
         targets (list): Targets to compile this predictor for. Pass `None` to compile for our default targets.
-        access (AccessMode): Predictor access.
         metadata (list): Metadata to use while compiling the function.
+        access (AccessMode): Predictor access.
         card (str | Path): Predictor card markdown string or path to card.
         media (Path): Predictor thumbnail image (jpeg or png) path.
         license (str): Predictor license URL. This is required for public predictors.
@@ -96,5 +111,5 @@ def compile (
         def wrapper (*args, **kwargs):
             return func(*args, **kwargs)
         wrapper.__predictor_spec = spec
-        return wrapper
+        return cast(Callable[P, R], wrapper)
     return decorator

{fxn-0.0.50 → fxn-0.0.52}/fxn/version.py

@@ -3,4 +3,4 @@
 #   Copyright © 2025 NatML Inc. All Rights Reserved.
 #
 
-__version__ = "0.0.50"
+__version__ = "0.0.52"

{fxn-0.0.50 → fxn-0.0.52}/fxn.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: fxn
-Version: 0.0.50
+Version: 0.0.52
 Summary: Run prediction functions locally in Python. Register at https://fxn.ai.
 Author-email: "NatML Inc." <hi@fxn.ai>
 License:                                  Apache License