Enable trainer tests (#495)

Author: AlbertvanHouten

.github/components-path-filters.yml

@@ -281,31 +281,30 @@ interactive_ai/workflows/geti_domain/train:
   - Makefile.shared-python
   - Makefile.shared
   - .github/workflows/component.yml
-# TODO: Re-enable CVS-166868
-#interactive_ai/workflows/otx_domain/trainer/otx_v2/xpu:
-#  - interactive_ai/workflows/otx_domain/trainer/otx_v2/xpu/**
-#  - interactive_ai/workflows/otx_domain/trainer/otx_v2/scripts/**
-#  - interactive_ai/workflows/otx_domain/trainer/otx_v2/tests/**
-#  - interactive_ai/workflows/otx_domain/trainer/otx_v2/download_pretrained_weights.py
-#  - interactive_ai/workflows/otx_domain/trainer/otx_v2/pyproject.toml
-#  - interactive_ai/workflows/otx_domain/trainer/otx_v2/run
-#  - libs/telemetry_tools/**
-#  - libs/kafka_tools/**
-#  - Makefile.shared-python
-#  - Makefile.shared
-#  - .github/workflows/component.yml
-# interactive_ai/workflows/otx_domain/trainer/otx_v2/gpu:
-#   - interactive_ai/workflows/otx_domain/trainer/otx_v2/gpu/**
-#   - interactive_ai/workflows/otx_domain/trainer/otx_v2/scripts/**
-#   - interactive_ai/workflows/otx_domain/trainer/otx_v2/tests/**
-#   - interactive_ai/workflows/otx_domain/trainer/otx_v2/download_pretrained_weights.py
-#   - interactive_ai/workflows/otx_domain/trainer/otx_v2/pyproject.toml
-#   - interactive_ai/workflows/otx_domain/trainer/otx_v2/run
-#   - libs/telemetry_tools/**
-#   - libs/kafka_tools/**
-#   - Makefile.shared-python
-#   - Makefile.shared
-#   - .github/workflows/component.yml
+interactive_ai/workflows/otx_domain/trainer/otx_v2/xpu:
+  - interactive_ai/workflows/otx_domain/trainer/otx_v2/xpu/**
+  - interactive_ai/workflows/otx_domain/trainer/otx_v2/scripts/**
+  - interactive_ai/workflows/otx_domain/trainer/otx_v2/tests/**
+  - interactive_ai/workflows/otx_domain/trainer/otx_v2/download_pretrained_weights.py
+  - interactive_ai/workflows/otx_domain/trainer/otx_v2/pyproject.toml
+  - interactive_ai/workflows/otx_domain/trainer/otx_v2/run
+  - libs/telemetry_tools/**
+  - libs/kafka_tools/**
+  - Makefile.shared-python
+  - Makefile.shared
+  - .github/workflows/component.yml
+interactive_ai/workflows/otx_domain/trainer/otx_v2/gpu:
+  - interactive_ai/workflows/otx_domain/trainer/otx_v2/gpu/**
+  - interactive_ai/workflows/otx_domain/trainer/otx_v2/scripts/**
+  - interactive_ai/workflows/otx_domain/trainer/otx_v2/tests/**
+  - interactive_ai/workflows/otx_domain/trainer/otx_v2/download_pretrained_weights.py
+  - interactive_ai/workflows/otx_domain/trainer/otx_v2/pyproject.toml
+  - interactive_ai/workflows/otx_domain/trainer/otx_v2/run
+  - libs/telemetry_tools/**
+  - libs/kafka_tools/**
+  - Makefile.shared-python
+  - Makefile.shared
+  - .github/workflows/component.yml
 
 # Web UI
 web_ui:

interactive_ai/workflows/otx_domain/trainer/otx_v2/gpu/Dockerfile

@@ -32,7 +32,7 @@ RUN apt update \
 	&& echo "/usr/local/nvidia/lib64" >> /etc/ld.so.conf.d/nvidia.conf
 
 # step 2: get python dependencies
-FROM base AS build
+FROM base AS runtime
 
 ENV UV_COMPILE_BYTECODE=1
 ENV UV_LINK_MODE=copy
@@ -44,6 +44,7 @@ ENV UV_PYTHON_DOWNLOADS=0
 COPY --link --from=libs . libs
 
 WORKDIR /interactive_ai/workflows/otx_domain/trainer/otx_v2
+RUN chown -R 10001:10001 /interactive_ai/workflows/otx_domain/trainer/otx_v2
 
 COPY --link --from=ghcr.io/astral-sh/uv:0.6.12@sha256:515b886e8eb99bcf9278776d8ea41eb4553a794195ef5803aa7ca6258653100d /uv /bin/uv
 
@@ -56,9 +57,6 @@ RUN --mount=type=cache,target=/root/.cache/uv \
     --mount=type=bind,source=pyproject.toml,target=pyproject.toml \
     uv sync --frozen --no-dev --no-editable
 
-# step 3: build final runtime image
-FROM base AS runtime
-
 # Install runtime dependencies
 RUN apt-get update && \
     apt-get install -y --no-install-recommends \
@@ -71,7 +69,6 @@ RUN apt-get update && \
     rm -rf /root/.cache/pip
 
 # Copy the application from the builder
-COPY --link --from=build --chown=10001 /interactive_ai/workflows/otx_domain/trainer/otx_v2 /interactive_ai/workflows/otx_domain/trainer/otx_v2
 COPY --link --from=cuda --chown=10001 /usr/local /usr/local
 
 # Place executables in the environment at the front of the path

interactive_ai/workflows/otx_domain/trainer/otx_v2/gpu/Makefile

@@ -26,10 +26,8 @@ style-fix: venv
 	uv run ruff check .. --config ../pyproject.toml --fix
 	uv run ruff format .. --config ../pyproject.toml
 
-# TODO CVS-166188 remove the '-' and 'true' workaround after fixing the tests
 test-unit: venv
-	-PYTHONPATH=../scripts uv run pytest ../tests --disable-warnings -v
-	true
+	PYTHONPATH=../scripts uv run pytest ../tests --disable-warnings -v
 
 test-component:
 	@echo "test-component not available"

interactive_ai/workflows/otx_domain/trainer/otx_v2/tests/conftest.py

@@ -33,3 +33,9 @@ def fxt_async_caller():
     AsyncCaller().start()
     yield
     AsyncCaller().close()
+
+
+@pytest.fixture(scope="session", autouse=True)
+def ensure_torch_hub_checkpoints():
+    checkpoints_dir = Path.home() / ".cache" / "torch" / "hub" / "checkpoints"
+    checkpoints_dir.mkdir(parents=True, exist_ok=True)

interactive_ai/workflows/otx_domain/trainer/otx_v2/xpu/Dockerfile

@@ -35,13 +35,11 @@ RUN apt-get purge -y --auto-remove gpg wget && \
     /tmp/* \
     /usr/share/man/* /usr/share/doc/*
 
-# step 2: get python dependencies
-FROM base AS build
-
 # Copy the service dependencies
 COPY --link --from=libs . libs
 
 WORKDIR /interactive_ai/workflows/otx_domain/trainer/otx_v2
+RUN chown -R 10001:10001 /interactive_ai/workflows/otx_domain/trainer/otx_v2
 
 COPY --link --from=ghcr.io/astral-sh/uv:0.6.12@sha256:515b886e8eb99bcf9278776d8ea41eb4553a794195ef5803aa7ca6258653100d /uv /bin/uv
 
@@ -56,12 +54,6 @@ RUN --mount=type=cache,target=/root/.cache/uv uv sync --frozen --no-dev --no-edi
 RUN uv pip uninstall torch
 RUN uv pip install torch==2.7.0 --default-index https://download.pytorch.org/whl/xpu
 
-# step 3: build final runtime image
-FROM base AS runtime
-
-# Copy the application from the builder
-COPY --link --from=build --chown=10001 /interactive_ai/workflows/otx_domain/trainer/otx_v2 /interactive_ai/workflows/otx_domain/trainer/otx_v2
-
 # Place executables in the environment at the front of the path
 ENV PATH="/interactive_ai/workflows/otx_domain/trainer/otx_v2/.venv/bin:/interactive_ai/workflows/otx_domain/trainer/otx_v2:$PATH"
 ENV PYTHONPATH="/interactive_ai/workflows/otx_domain/trainer/otx_v2"
@@ -70,4 +62,4 @@ ENV HF_HUB_OFFLINE=1
 USER non-root
 
 WORKDIR /home/non-root
-WORKDIR /interactive_ai/workflows/otx_domain/trainer/otx_v2
+WORKDIR /interactive_ai/workflows/otx_domain/trainer/otx_v2

interactive_ai/workflows/otx_domain/trainer/otx_v2/xpu/Makefile

@@ -26,10 +26,8 @@ style-fix: venv
 	uv run ruff check .. --config ../pyproject.toml --fix
 	uv run ruff format .. --config ../pyproject.toml
 
-# TODO CVS-166188 remove the '-' and 'true' workaround after fixing the tests
 test-unit: venv
-	-PYTHONPATH=../scripts uv run pytest ../tests --disable-warnings -v
-	true
+	PYTHONPATH=../scripts uv run pytest ../tests --disable-warnings -v
 
 test-component:
 	@echo "test-component not available"