Merge pull request #23 from BeastByteAI/phonnx

Phonnx

.github/workflows/pypi-deploy.yml

@@ -0,0 +1,31 @@
+name: PyPi Deploy
+
+on:
+  release:
+    types: [published]
+  workflow_dispatch:
+
+jobs:
+  deploy:
+    runs-on: ubuntu-latest
+    steps:
+    - name: Checkout code
+      uses: actions/checkout@v3
+
+    - name: Setup Python
+      uses: actions/setup-python@v4
+      with:
+        python-version: '3.10'
+
+    - name: Install dependencies
+      run: |
+        python -m pip install --upgrade pip
+        pip install build twine
+
+    - name: Build and publish
+      env:
+        TWINE_USERNAME: __token__
+        TWINE_PASSWORD: ${{ secrets.PYPI_TOKEN }}
+      run: |
+        python -m build
+        twine upload dist/*

falcon/runtime.py

@@ -1,30 +1,22 @@
 from typing import Any, Union, Dict, List, Type, Optional
+
 try:
-    import onnxruntime as ort
+    from phonnx.runtime import Runtime as _PhonnxRuntime
 except (ImportError, ModuleNotFoundError):
-    print("ONNXRuntime is not installed. Inference modules will not work.")
-    ort = None
+    print("ONNXRuntime/PHONNX is not installed. Inference modules will not work.")
+    _PhonnxRuntime = None
 import numpy as np
-from abc import ABC, abstractmethod
-
 
-class BaseRuntime(ABC):
-    @abstractmethod
-    def run(self, X: np.ndarray, **kwargs: Any) -> Any:
-        pass
 
-
-class ONNXRuntime(BaseRuntime):
+class ONNXRuntime:
     """
-    Runtime for ONNX models. This runtime can only run onnx models produced by falcon.
+    Runtime for ONNX models based on PHONNX.
     """
 
     def __init__(self, model: Union[bytes, str]):
-        if ort is None:
-            raise RuntimeError(
-                "ONNXRuntime is not installed. Please install it with `pip install onnxruntime`."
-            )
-        self.ort_session = ort.InferenceSession(model)
+        if _PhonnxRuntime is None:
+            raise ImportError("PHONNX is not installed.")
+        self.runtime = _PhonnxRuntime(model=model)
 
     def run(
         self, X: np.ndarray, outputs: str = "final", **kwargs: Any
@@ -49,50 +41,4 @@ def run(
                 f"Expected `outputs` to be one of [all, final], got `{outputs}`."
             )
 
-        inputs = self._get_inputs(X)
-        output_names = self._get_output_names(outputs)
-
-        return self.ort_session.run(output_names, inputs)
-
-    def _get_inputs(self, X: np.ndarray) -> Dict:
-        ort_inputs = self.ort_session.get_inputs()
-        inputs = {}
-
-        for i, inp in enumerate(ort_inputs):
-            dtype: Any
-            if str(inp.type) == "tensor(float)":
-                dtype = np.float32
-            elif str(inp.type) == "tensor(string)":
-                dtype = np.str_
-            elif "int" in str(inp.type).lower():
-                dtype = np.int64
-            else:
-                RuntimeError(
-                    f"The model input type should be one of [str, int, float], got f{inp.type}"
-                )
-            if len(ort_inputs) > 1:
-                inputs[str(inp.name)] = np.expand_dims(X[:, i], 1).astype(dtype)
-            else:
-                inputs[str(inp.name)] = X.astype(dtype)
-        return inputs
-
-    def _get_output_names(self, outputs: str = "final") -> List[str]:
-        ort_outputs = self.ort_session.get_outputs()
-        output_names = [o.name for o in ort_outputs]
-        if outputs == "final" and len(ort_outputs) > 1:
-            idx_ = []
-            for name in output_names:
-                if not name[0:10] in ("falcon_pl_", "falcon-pl-"):
-                    raise RuntimeError("One of the output nodes has an invalid name.")
-                idx = int(name.split("/")[0][10:])
-                idx_.append(idx)
-            max_idx = max(idx_)
-            output_names = [
-                n
-                for n in output_names
-                if (
-                    n.startswith(f"falcon_pl_{max_idx}")
-                    or n.startswith(f"falcon-pl-{max_idx}")
-                )
-            ]
-        return output_names
+        return self.runtime.run(X, outputs_to_return=outputs)

pyproject.toml

@@ -14,10 +14,11 @@ dependencies = [
 "imbalanced-learn>=0.8.1",
 "pyarrow>=8.0.0",
 "optuna>=3.0.0", 
-"packaging>=20.0.0"
+"packaging>=20.0.0",
+"phonnx>=0.0.1",
 ]
 name = "falcon-ml"
-version = "0.6.0"
+version = "0.7.0"
 authors = [
   { name="Oleg Kostromin", email="[email protected]" },
   { name="Iryna Kondrashchenko", email="[email protected]" },
@@ -26,7 +27,7 @@ authors = [
 description = "AutoML library for fast experementations."
 readme = "README.md"
 license = {text = "MIT"}
-requires-python = ">=3.8"
+requires-python = ">=3.9"
 classifiers = [
     "Programming Language :: Python :: 3",
     "License :: OSI Approved :: MIT License",