WIP.

Author: lc525

docs/examples/sklearn/infer.py

@@ -0,0 +1,37 @@
+import requests
+from sklearn import datasets, svm, metrics
+from sklearn.model_selection import train_test_split
+
+# The digits dataset
+digits = datasets.load_digits()
+
+# To apply a classifier on this data, we need to flatten the image, to
+# turn the data in a (samples, feature) matrix:
+n_samples = len(digits.images)
+data = digits.images.reshape((n_samples, -1))
+
+# Create a classifier: a support vector classifier
+classifier = svm.SVC(gamma=0.001)
+
+# Split data into train and test subsets
+X_train, X_test, y_train, y_test = train_test_split(
+    data, digits.target, test_size=0.5, shuffle=False)
+
+
+x_0 = X_test[0:1]
+inference_request = {
+    "inputs": [
+        {
+          "name": "predict",
+          "shape": x_0.shape,
+          "datatype": "FP32",
+          "data": x_0.tolist()
+        }
+    ]
+}
+print(inference_request)
+
+endpoint = "http://localhost:8080/v2/models/mnist-svm/versions/v0.1.0/infer"
+response = requests.post(endpoint, json=inference_request)
+
+print(response.json())

mlserver/metrics/prometheus.py

@@ -31,14 +31,21 @@ def configure_metrics(settings: Settings):
     values.ValueClass = values.get_value_class()
 
 
+async def try_remove(filename: str):
+    try:
+        os.remove(filename)
+    except FileNotFoundError:
+        pass
+
 async def stop_metrics(settings: Settings, pid: int):
     if not settings.parallel_workers:
         return
 
     mark_process_dead(pid)
     pattern = os.path.join(settings.metrics_dir, f"*_{pid}.db")
     matching_files = glob.glob(pattern)
-    await asyncio.gather(*[remove(f) for f in matching_files])
+    if len(matching_files) > 0:
+        await asyncio.gather(*[try_remove(f) for f in matching_files])
 
 
 class PrometheusEndpoint:

mlserver/settings.py

@@ -165,7 +165,7 @@ class Settings(BaseSettings):
         extra="ignore",
     )
 
-    debug: bool = True
+    debug: bool = False
 
     parallel_workers: int = DEFAULT_PARALLEL_WORKERS
     """When parallel inference is enabled, number of workers to run inference

mlserver/utils.py

@@ -78,31 +78,6 @@ def event_loop_policy(self) -> asyncio.AbstractEventLoopPolicy:
         else:
             return asyncio.DefaultEventLoopPolicy
 
-    def install_uvloop_when_available(self):
-        """
-        In python versions < 3.16, install uvloop as the default event loop policy.
-        Deprecated from python 3.12, but used within pytest tests until pytest itself provides
-        an alternative way to manage event loops across different python versions.
-        """
-        python_version_info = sys.version_info[:2]
-        if python_version_info >= (3, 12):
-            warnings.warn(
-                'AsyncManager::install() is deprecated in favor of AsyncManager::run() '
-                'for Python >= 3.12.',
-                DeprecationWarning,
-                stacklevel=1,
-            )
-        if python_version_info < (3, 16):
-            if type(asyncio.get_event_loop_policy()).__module__.startswith("uvloop"):
-                return
-            if self._event_loop_backend == EventLoopBackend.UVLOOP:
-                import uvloop
-                uvloop.install()
-        else:
-            raise RuntimeError(
-                "uvloop install via EventLoopPolicy is only supported for Python < 3.16"
-            )
-
     def run(self,
             coro: Coroutine[Any, Any, _CoroRetT], *,
             debug: Optional[bool] = None) -> _CoroRetT:

poetry.lock

@@ -127,7 +127,7 @@ aiohttp = "3.12.8"
 aiohttp-retry = "2.9.1"
 ## Used for FastAPI Async testing
 httpx = "0.27.0"
-kafka-python-ng = "2.2.3"
+kafka-python = "2.2.15"
 tenacity = "8.4.1"
 pyyaml = "6.0.1"
 conda-pack = "0.7.1"
@@ -188,7 +188,7 @@ torch = "^2.4"
 pytorch-lightning = "^2.4"
 torchmetrics = "1.6.0"
 torchvision = "0.19.1"
-mlflow = "2.19.0"
+mlflow = ">=3.0.0"
 
 ## Dev dependencies from HuggingFace
 # TODO: Relax when we deprecate Conversation pipeline

pyproject.toml

@@ -1,6 +1,5 @@
 import pytest
 import os
-import asyncio
 import numpy as np
 import tensorflow as tf
 
@@ -13,7 +12,6 @@
 from mlserver.context import model_context
 from mlserver.settings import ModelSettings, ModelParameters
 from mlserver.types import InferenceRequest
-from mlserver.utils import AsyncManager
 
 from mlserver_alibi_detect import AlibiDetectRuntime
 
@@ -26,16 +24,6 @@
 TESTS_PATH = os.path.dirname(__file__)
 TESTDATA_PATH = os.path.join(TESTS_PATH, "testdata")
 
-
-# @pytest.fixture(autouse=True)
-# def event_loop():
-#     # By default use uvloop for tests
-#     AsyncManager().install_uvloop()
-#     loop = asyncio.get_event_loop()
-#     yield loop
-#     loop.close()
-
-
 @pytest.fixture
 def inference_request() -> InferenceRequest:
     payload_path = os.path.join(TESTDATA_PATH, "inference-request.json")