fix(deps): update machine-learning (#10740)

* fix(deps): update machine-learning

* update openvino options, cuda

* update openvino build

* fix indentation

* update minimum nvidia driver

---------

Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>
Co-authored-by: mertalev <[email protected]>

docker/docker-compose.prod.yml

@@ -33,6 +33,8 @@ services:
       dockerfile: Dockerfile
       args:
         - DEVICE=cpu # set to one of [armnn, cuda, openvino, openvino-wsl] for accelerated inference
+    ports:
+      - 3003:3003
     volumes:
       - model-cache:/cache
     env_file:

docs/docs/features/ml-hardware-acceleration.md

@@ -38,7 +38,7 @@ You do not need to redo any machine learning jobs after enabling hardware accele
 
 - The GPU must have compute capability 5.2 or greater.
 - The server must have the official NVIDIA driver installed.
-- The installed driver must be >= 535 (it must support CUDA 12.2).
+- The installed driver must be >= 545 (it must support CUDA 12.3.2).
 - On Linux (except for WSL2), you also need to have [NVIDIA Container Toolkit][nvct] installed.
 
 #### OpenVINO

machine-learning/Dockerfile

@@ -2,9 +2,7 @@ ARG DEVICE=cpu
 
 FROM python:3.11-bookworm@sha256:7bec1574675e7fd9e3a540a03cd7d6811c59ca261bd300cd665369d8f435298a as builder-cpu
 
-FROM openvino/ubuntu22_runtime:2023.3.0@sha256:176646df619032ea6c10faf842867119c393e7497b7f88b5e307e932a0fd5aa8 as builder-openvino
-USER root
-RUN apt-get update && apt-get install -y --no-install-recommends python3-dev
+FROM builder-cpu as builder-openvino
 
 FROM builder-cpu as builder-cuda
 
@@ -38,13 +36,15 @@ RUN poetry install --sync --no-interaction --no-ansi --no-root --with ${DEVICE}
 
 FROM python:3.11-slim-bookworm@sha256:17ec9dc2367aa748559d0212f34665ec4df801129de32db705ea34654b5bc77a as prod-cpu
 
-FROM openvino/ubuntu22_runtime:2023.3.0@sha256:176646df619032ea6c10faf842867119c393e7497b7f88b5e307e932a0fd5aa8 as prod-openvino
-USER root
-# TODO: remove this once the image has the fix for https://github.com/intel/compute-runtime/issues/710
-ENV NEOReadDebugKeys=1 \
-    OverrideGpuAddressSpace=48
+FROM prod-cpu as prod-openvino
 
-FROM nvidia/cuda:12.2.2-cudnn8-runtime-ubuntu22.04@sha256:2d913b09e6be8387e1a10976933642c73c840c0b735f0bf3c28d97fc9bc422e0 as prod-cuda
+COPY scripts/configure-apt.sh ./
+RUN ./configure-apt.sh && \
+    apt-get update && \
+    apt-get install -t unstable --no-install-recommends -yqq intel-opencl-icd && \
+    rm configure-apt.sh
+
+FROM nvidia/cuda:12.3.2-cudnn9-runtime-ubuntu22.04@sha256:fa44193567d1908f7ca1f3abf8623ce9c63bc8cba7bcfdb32702eb04d326f7a8 as prod-cuda
 
 COPY --from=builder-cuda /usr/local/bin/python3 /usr/local/bin/python3
 COPY --from=builder-cuda /usr/local/lib/python3.11 /usr/local/lib/python3.11
@@ -71,13 +71,15 @@ COPY --from=builder-armnn \
     /opt/armnn/
 
 FROM prod-${DEVICE} as prod
+ARG DEVICE
 
 RUN apt-get update && \
-    apt-get install -y --no-install-recommends tini libmimalloc2.0 && \
+    apt-get install -y --no-install-recommends tini $(if ! [ "$DEVICE" = "openvino" ]; then echo "libmimalloc2.0"; fi) && \
+    apt-get autoremove -yqq && \
+    apt-get clean && \
     rm -rf /var/lib/apt/lists/*
 
 WORKDIR /usr/src/app
-ARG DEVICE
 ENV TRANSFORMERS_CACHE=/cache \
     PYTHONDONTWRITEBYTECODE=1 \
     PYTHONUNBUFFERED=1 \

machine-learning/app/models/transforms.py

@@ -47,8 +47,8 @@ def pil_to_cv2(image: Image.Image) -> NDArray[np.uint8]:
 def decode_pil(image_bytes: bytes | IO[bytes] | Image.Image) -> Image.Image:
     if isinstance(image_bytes, Image.Image):
         return image_bytes
-    image = Image.open(BytesIO(image_bytes) if isinstance(image_bytes, bytes) else image_bytes)
-    image.load()  # type: ignore
+    image: Image.Image = Image.open(BytesIO(image_bytes) if isinstance(image_bytes, bytes) else image_bytes)
+    image.load()
     if not image.mode == "RGB":
         image = image.convert("RGB")
     return image

machine-learning/app/sessions/ort.py

@@ -83,17 +83,21 @@ def provider_options(self, provider_options: list[dict[str, Any]]) -> None:
 
     @property
     def _provider_options_default(self) -> list[dict[str, Any]]:
-        options = []
+        provider_options = []
         for provider in self.providers:
             match provider:
                 case "CPUExecutionProvider" | "CUDAExecutionProvider":
-                    option = {"arena_extend_strategy": "kSameAsRequested"}
+                    options = {"arena_extend_strategy": "kSameAsRequested"}
                 case "OpenVINOExecutionProvider":
-                    option = {"device_type": "GPU_FP32", "cache_dir": (self.model_path.parent / "openvino").as_posix()}
+                    options = {
+                        "device_type": "GPU",
+                        "precision": "FP32",
+                        "cache_dir": (self.model_path.parent / "openvino").as_posix(),
+                    }
                 case _:
-                    option = {}
-            options.append(option)
-        return options
+                    options = {}
+            provider_options.append(options)
+        return provider_options
 
     @property
     def sess_options(self) -> ort.SessionOptions:

machine-learning/app/test_main.py

@@ -212,7 +212,7 @@ def test_sets_default_provider_options(self, ov_device_ids: list[str]) -> None:
         session = OrtSession(model_path, providers=["OpenVINOExecutionProvider", "CPUExecutionProvider"])
 
         assert session.provider_options == [
-            {"device_type": "GPU_FP32", "cache_dir": "/cache/ViT-B-32__openai/openvino"},
+            {"device_type": "GPU", "precision": "FP32", "cache_dir": "/cache/ViT-B-32__openai/openvino"},
             {"arena_extend_strategy": "kSameAsRequested"},
         ]